RTP and RTCP (RFC 3550, RFC 8834)

View Source

This page tracks the RTP and RTCP requirements implemented under src/rtp and src/cc: the RTP and RTCP codecs, header extensions, feedback messages, receiver statistics, RTCP timing, retransmission, the payload helpers, and congestion control (bandwidth estimation, pacing, circuit breakers). Read it when you need to know whether a packet type or rule is supported before relying on it. Scheduling decisions (when to send a report, which feedback to merge) belong to the transport process and are marked as not implemented here until that process exists. Test names refer to test/<module>_tests.erl and to the properties in test/prop_webrtc_rtp.erl and test/prop_webrtc_cc.erl.

RFC 3550: RTP data packets

IDSectionRequirementModuleTestStatus
RTP-15.1Fixed header: version 2, P, X, CC, M, PT, sequence, timestamp, SSRCwebrtc_rtpwebrtcrtp_tests:minimal_test, bad_version_test, prop_rtp_roundtripUnit tested
RTP-25.1CSRC list of CC (0..15) entrieswebrtc_rtpcsrc_and_marker_testUnit tested
RTP-35.1, A.1Padding: last octet counts the padding including itself, at least 1 and no more than the octets after the headerwebrtc_rtppaddingtest, padding_only_test, bad_padding_testUnit tested
RTP-45.3.1, A.1Header extension: 16-bit profile, length in 32-bit words, length checked against the packetwebrtc_rtpextensionlength_test, max_extension_length_testUnit tested
RTP-55.3.1Unknown extension profiles are carried and can be ignoredwebrtc_rtpunknown_profile_kept_raw_testUnit tested
RTP-6A.1Truncated or malformed packets are rejected, never crashwebrtc_rtptruncation_test, prop_rtp_decode_never_crashes, prop_rtp_mutation_never_crashesUnit tested
RTP-75.1Sequence numbers and timestamps wrap modulo 2^16 and 2^32webrtc_rtp_seqwebrtcrtp_seq_tests:seq_test, tstest, prop_seq_unwrapUnit tested
RTP-85.1Header fields rewritten in place for forwardingwebrtc_rtppeek_rewrite_testUnit tested
RTP-9A.1Source validation: probation of MIN_SEQUENTIAL packets, MAX_DROPOUT, MAX_MISORDER, resync after two sequential packets, cycle countwebrtc_rtp_statsprobation_test, probation_restart_test, wraparound_test, misorder_and_dropout_testUnit tested
RTP-10A.3Expected, cumulative lost clamped to 24-bit signed, fraction lost per intervalwebrtc_rtp_statsloss_and_fraction_test, duplicates_negative_loss_test, lost_clamp_testUnit tested
RTP-116.4.1, A.8Interarrival jitter in timestamp units, robust to timestamp wrapwebrtc_rtp_statsjitter_test, jitter_ts_wrap_test, jitter_time_unit_testUnit tested
RTP-126.4.1Sender packet and payload octet counts for SRwebrtc_rtp_statssender_testUnit tested
RTP-138.1, 8.2Random SSRC choice and collision detection: our SSRCs avoid those of the remote description; a peer packet or RTCP sender using one of ours makes that sender take a new SSRC, send a BYE for the old one and report {ssrc_changed, Mid, Old, New}, later descriptions carrying the new onewebrtc_jsep, webrtc_media_out, webrtc_media_in, webrtc_pcwebrtc_jsep_tests:ssrc_clash_avoided_test, webrtc_media_tests:ssrc_collision_testUnit tested

RFC 3550: RTCP packets

IDSectionRequirementModuleTestStatus
RTCP-16.4.1SR: sender info, up to 31 report blocks, profile-specific extension keptwebrtc_rtcpwebrtc_rtcp_tests:sr_sdes_test, sr_with_reports_and_extension_testUnit tested
RTCP-26.4.2RR with report blocks; cumulative lost is signed 24-bitwebrtc_rtcprr_remb_test, negative_lost_testUnit tested
RTCP-36.4.1LSR / DLSR and round-trip time A - LSR - DLSR (RFC example: 6.125 s)webrtc_rtp_statslsr_dlsr_test, rtt_rfc_example_test, rtt_wrap_testUnit tested
RTCP-46.5SDES chunks: items, null terminator, 32-bit padding, CNAME, PRIV prefixwebrtc_rtcpsdesitems_test, sdes_empty_chunk_layout_test, sdes_errors_testUnit tested
RTCP-56.6BYE with SSRC list and optional reason padded with null octetswebrtc_rtcpbye_testUnit tested
RTCP-66.7APP with subtype, 4-octet name and datawebrtc_rtcpapp_testUnit tested
RTCP-76.1, A.2Version 2 on every packet; lengths add up to the datagram sizewebrtc_rtcpcompoundvalidation_test, truncationtestUnit tested
RTCP-86.1, A.2Compound check (mode => compound): first packet SR or RR, padding only on the last packetwebrtc_rtcpcompoundvalidation_testUnit tested
RTCP-96.1Compound packet carries an SDES CNAME (validate_compound/1)webrtc_rtcpsr_sdes_test, rr_remb_test, has_cname_testUnit tested
RTCP-106.4.1, A.2Padding count valid (1..length)webrtc_rtcppaddingtestUnit tested
RTCP-11A.2Unknown packet types kept as raw, the others still decodedwebrtc_rtcpunknown_type_preserved_test, prop_rtcp_roundtripUnit tested
RTCP-126.1Truncated or corrupted RTCP returns an error, never crasheswebrtc_rtcptruncationtest, prop_rtcp_decode_never_crashes, prop_rtcp_mutation_never_crashesUnit tested
RTCP-136.4.2More than 31 sources need several RR packets: the first SR or RR carries 31 blocks, extra RRs the rest, within the RTCP budget and rotating the sourceswebrtc_media_outwebrtc_rtcp_tests:sr_with_reports_and_extension_test (32 blocks in one packet rejected), webrtc_media_tests:many_sources_reported_test (80 sources)Unit tested
RTCP-146.2, 6.3.1, A.7Interval: senders / receivers bandwidth split, Tmin, randomisation in [0.5, 1.5), e - 3/2 compensationwebrtc_rtcp_intervalwebrtcrtcp_interval_tests:deterministic_test, randomisedtestUnit tested
RTCP-156.2, A.7Initial interval uses half the minimumwebrtc_rtcp_intervalmininterval_testUnit tested
RTCP-166.3.3avg_rtcp_size moving average (1/16)webrtc_rtcp_intervalavpfhelpers_testUnit tested
RTCP-176.3.5Participant timeout M * Td, M = 5webrtc_rtcp_intervaltimeouttestUnit tested
RTCP-186.3.6Timer reconsideration on expirywebrtc_rtcp_intervalreconsidertestUnit tested
RTCP-196.3.4Reverse reconsideration after BYEwebrtc_rtcp_intervalreversereconsider_testUnit tested
RTCP-206.3, A.7RTCP scheduler state: tp, tn, average compound size (lower layers counted), members and senders from the streams, we_sent, RTCP bandwidth 5% of the session (both directions of the send estimate), interval reconsidered at expiry and moved closer when members leavewebrtc_rtcp_sched, webrtc_mediawebrtc_rtcp_sched_tests, webrtc_media_tests:many_sources_reported_testUnit tested; the BYE back-off of §6.3.7 is not needed point to point (we send no BYE flood)
RTCP-216.4, 6.4.2Report blocks for every received source: 31 in the first SR or RR, more in RR packets that follow it, each source once per compound, the compound kept under 1150 bytes before SRTCP; when not all fit, the first source reported rotates and a source's interval statistics only advance when it is reportedwebrtc_media_outwebrtc_media_tests:many_sources_reported_testUnit tested

RFC 8834: RTP usage in WebRTC

IDSectionRequirementModuleTestStatus
WRTP-14.1Send and receive SR, RR, SDES and BYEwebrtc_rtcpwebrtc_rtcp_tests:sr_sdes_test, rr_remb_test, bye_testUnit tested
WRTP-24.1RTCP SDES MID item (RFC 9143 §15.1)webrtc_rtcpsdes_items_testUnit tested
WRTP-34.1Reception of CSRC listswebrtc_rtpcsrc_and_marker_testUnit tested
WRTP-44.1Reduced minimum interval 360 / kbps (RFC 3550 §6.2)webrtc_rtcp_intervalmininterval_testUnit tested
WRTP-54.1Non-reduced minimum used for the participant timeoutwebrtc_rtcp_intervaltimeouttestUnit tested
WRTP-64.1Timer reconsideration and reverse reconsiderationwebrtc_rtcp_intervalreconsidertest, reversereconsider_testUnit tested
WRTP-74.1Ignore unknown RTCP packet types and header extensions while processing the known oneswebrtc_rtcp, webrtc_rtpunknown_type_preserved_test, unknown_profile_kept_raw_testUnit tested
WRTP-84.5RTP and RTCP multiplexed on one flow (RFC 5761)webrtc_rtpclassifytestUnit tested
WRTP-94.6Receive and send non-compound RTCP (RFC 5506)webrtc_rtcpcompoundvalidation_testUnit tested
WRTP-105.1.1Understand FIRwebrtc_rtcp_fbwebrtc_rtcp_fb_tests:fir_testUnit tested
WRTP-115.1.2Understand PLIwebrtc_rtcp_fbpli_testUnit tested
WRTP-125.1.3SLI: a sender acts on it, a receiver generates it when its codec has sliceswebrtc_rtcp_fb, webrtc_media_fbsli_test, webrtc_media_loss_tests, webrtc_pc_SUITE:loss_indications, webrtc_pc_interop_SUITE:slice_loss_indicationsInteroperability tested with Pion both ways. The library has no codec: the owner gets {slice_loss, ...} and the encoding plugin feedback/2; the application (send_slice_loss/3,4) or the decoding plugin generates it
WRTP-135.1.4RPSI: a sender acts on it, a receiver generates it on lost decoder synchronisationwebrtc_rtcp_fb, webrtc_media_fbrpsilibwebrtc_test, rpsi_padding_test, rpsi_errors_test, prop_rtcp_roundtrip, webrtc_pc_SUITE:loss_indicationsUnit and e2e tested, the codec against libwebrtc's former RPSI test packet. Events, plugin and API as for SLI (send_reference_picture/4,5). No interop peer implements RPSI
WRTP-245.1SLI and RPSI used only when nack sli / nack rpsi is negotiated; we offer both on videowebrtc_jsep_caps, webrtc_media_fbignored_test, not_negotiated_testUnit tested
WRTP-145.1.6Understand TMMBR, answer with TMMBN, and follow the limit as media senderwebrtc_rtcp_fb, webrtc_tmmbr, webrtc_media_limitwebrtc_media_tmmbr_tests, webrtc_pc_SUITE:tmmbr_limits_senderUnit and e2e tested (Erlang to Erlang: browsers do not negotiate ccm tmmbr)
WRTP-155.2Header extensions follow RFC 8285webrtc_rtp_extwebrtc_rtp_ext_testsUnit tested
WRTP-165.2.1Rapid synchronisation (RFC 6051) in addition to SR, negotiated: ntp-64 and ntp-56 offeredwebrtc_media_sync, webrtc_jsep_capswebrtc_media_sync_tests, webrtc_sync_gst_SUITE, webrtc_pc_interop_SUITE:sr_on_requestInteroperability tested: the extension both ways with GStreamer's rtphdrextntp64, RTCP-SR-REQ with Pion
SYNC-16051 3.3ntp-64 from the clock and mapping of our SR (the wall clock when the packet is made, the point SR extrapolate from)webrtc_media_syncmapping_testUnit tested: the receiver's mapping from the extension and from the next SR agree within 5 ms
SYNC-26051 3.3Sent on a video random access point and the associated audio packets; we also send it until the peer has an SRwebrtc_media_syncinclusion_testUnit tested (first packet of each keyframe, the next audio packet, every packet until a report block has an LSR)
SYNC-36051 3.3ntp-56: the upper 8 bits of seconds from SRwebrtc_media_syncntp56_testUnit tested (received only; we send ntp-64)
SYNC-46051 3.3Regular SR still sentwebrtc_media_outwebrtc_media_testsUnit tested
SYNC-56051 3.2RTCP-SR-REQ: RTPFB FMT 5, no FCI, length 2webrtc_rtcp_fbsrreq_test, webrtc_corpus_tests (rtcp/6.rtcp refused)Unit tested
SYNC-66051 3.2Media source answers SR-REQ with an SR as soon as the early feedback rules allowwebrtc_media_sync, webrtc_media_outsrreq_test, webrtc_pc_interop_SUITE:sr_on_requestInteroperability tested with Pion (pion/rtcp RapidResynchronizationRequest)
SYNC-76051 3.2Receiver unable to synchronise sends SR-REQ, at most once per reporting intervalwebrtc_media_syncsrreq_testUnit tested (after 1 s of media without a mapping, with the regular report, at most four per report)
WRTP-175.2.2Client-to-mixer audio level (RFC 6464), sent with the audio_level option of send_rtp, encrypted with RFC 6904 when negotiated (the capability is required, its use recommended)webrtc_rtp_ext, webrtc_media_outaudio_level_test, webrtc_pc_SUITE:header_extension_encryption, webrtc_pc_interop_SUITE:erlang_sends_audio_level, peer_sends_audio_levelInteroperability tested in clear (Chromium and Firefox read our level, we read theirs); encrypted between Erlang peers
WRTP-185.2.4MID header extensionwebrtc_rtp_extsdes_testUnit tested
WRTP-195.2.5Video orientation (CVO) header extensionwebrtc_rtp_extvideo_orientation_testUnit tested
WRTP-206.1Senders understand generic NACKwebrtc_rtcp_fb, webrtc_nack_buffernack_test, webrtc_nack_buffer_tests:buffer_testUnit tested
WRTP-216.1Receivers support RTX with SSRC multiplexing (RFC 4588)webrtc_rtxwebrtc_rtx_tests:rtx_testUnit tested
WRTP-226.1RTT estimate from SR / RR to decide on retransmission: each report block's RTT paces the NACK requests we repeat and keeps a packet from being resent twice within one RTTwebrtc_rtp_stats, webrtc_media_in, webrtc_nack, webrtc_nack_bufferrtt_rfc_example_test, buffer_test, webrtc_media_rtt_testsUnit tested
WRTP-23RFC 3550 §6.6, §6.3.5Receive streams end with a BYE or after 60 s without packets: receiver, NACK, rid, RTX and SRTP state removed; at most 16 receive streams per transceiver and 256 per session, packets of further SSRCs dropped (too_many_streams); SSRCs learned from packets kept across renegotiation only while livewebrtc_media_in, webrtc_mediawebrtc_media_tests:ssrc_churn_bounded_test, bye_removes_stream_testUnit tested

RFC 4585: RTP/AVPF

IDSectionRequirementModuleTestStatus
AVPF-16.1Feedback header: FMT, PT 205 / 206, sender and media SSRCwebrtc_rtcp_fbnackerrors_test, prop_rtcp_roundtripUnit tested
AVPF-26.2.1Generic NACK: one or more PID / BLP itemswebrtc_rtcp_fbnacktest, nack_errors_testUnit tested
AVPF-36.2.1Conversion between lost sequence numbers and PID / BLP, across wraparoundwebrtc_rtcp_fbnackhelpers_test, prop_nack_itemsUnit tested
AVPF-46.3.1PLI has no FCI (length 2)webrtc_rtcp_fbpli_testUnit tested
AVPF-56.3.2SLI: First, Number, PictureIDwebrtc_rtcp_fbsli_testUnit tested
AVPF-66.3.3RPSI: PB, payload type, native bit string, padded to 32 bitswebrtc_rtcp_fbrpsilibwebrtc_test, rpsi_padding_test, rpsi_errors_test, prop_rtcp_roundtripUnit tested (use in WRTP-13)
AVPF-76.4Application layer feedback other than REMB kept rawwebrtc_rtcp_fbafb_other_raw_testUnit tested
AVPF-83.4 d, 3.5.1Initial Tmin 0 point-to-point, 1 s multipartywebrtc_rtcp_intervalmininterval_testUnit tested
AVPF-93.5.3Tmin is 0 after the first regular reportwebrtc_rtcp_intervalmininterval_testUnit tested
AVPF-103.5.2T_dither_max: 0 point-to-point, l * T_rr with l = 0.5 otherwisewebrtc_rtcp_intervalavpfhelpers_testUnit tested
AVPF-113.5.3T_rr_current_interval = RND * T_rr_interval, RND in [0.5, 1.5)webrtc_rtcp_intervalavpfhelpers_testUnit tested
AVPF-123.5.3Timeout uses T_rr_interval instead of Tmin when setwebrtc_rtcp_intervaltimeouttestUnit tested
AVPF-133.5.2, 3.5.3Feedback point to point (T_dither_max 0): sent at once while the RTCP bandwidth allows (budget of one second of it, at least one datagram), otherwise kept (at most 32 packets) until it does or merged into a regular report that comes first; regular reports no closer than the local T_rr_interval (rtcp_report_interval, default 1 s), the first one after the AVPF initial intervalwebrtc_rtcp_sched, webrtc_media_outwebrtc_rtcp_sched_tests:feedback_budget_test, report_interval_floor_test, first_report_early_testUnit tested: feedback over the budget waits for the budget rather than for the next regular report as early mode does, which would make retransmission requests and congestion feedback too late; feedback suppression of multiparty sessions is not implemented

RFC 5104: codec control messages

IDSectionRequirementModuleTestStatus
CCM-14.2.1TMMBR FCI: SSRC, 6-bit exponent, 17-bit mantissa, 9-bit overheadwebrtc_rtcp_fbtmmbr_testUnit tested
CCM-24.2.1.1Exponent and mantissa chosen so the encoded rate never exceeds the requested onewebrtc_rtcp_fbbitrate_precision_testUnit tested
CCM-34.2.2TMMBN, including an empty bounding setwebrtc_rtcp_fbtmmbn_testUnit tested
CCM-44.3.1FIR: media SSRC 0, FCI SSRC and sequence number, reserved bits ignoredwebrtc_rtcp_fbfir_testUnit tested
CCM-53.5.4.2TMMBR bounding set computationwebrtc_tmmbrwebrtc_tmmbr_tests, prop_bounding_set_envelope, prop_bounding_set_minimalUnit tested; the properties check the set against the definition (same lower envelope as every tuple, no member superfluous), exactly with rationals
CCM-74.2.1.2Media sender: new tuples merged with the previous bounding set, the owner's tuple replacedwebrtc_tmmbrnon_owner_test, owner_replaces_testUnit tested
CCM-84.2.1.2A higher limit applied only after 2 x RTT + T_dither_max; a lower one at oncewebrtc_tmmbr, webrtc_media_limithold_off_test, bye_lifts_limit_testUnit tested
CCM-94.2.1.2Owners that leave (BYE) or time out removed, with a new TMMBNwebrtc_tmmbr, webrtc_media_limitleave_test, expire_test, bye_lifts_limit_testUnit tested (timeout 25 s, five regular intervals of at least 5 s)
CCM-104.2.2.2A TMMBN answers every TMMBR, even one that leaves the set unchanged; media SSRC 0webrtc_media_limitrequest_answered_testUnit tested
CCM-113.5.4.1Sender stays within the feasible region: net media rate at the stream's packet ratewebrtc_rate_limit, webrtc_media_limitprop_rate_limit_bound, limit_followed_test, zero_holds_testUnit tested (packets wait in a token bucket, the oldest dropped when it is full; FEC counts against it; no probes on the stream)
CCM-124.2.1.2Media receiver: TMMBR repeated until a TMMBN lists it or already limits morewebrtc_media_limitrequest_answered_testUnit tested
CCM-137Used only when a=rtcp-fb:* ccm tmmbr is negotiated; answered without smaxprwebrtc_jsep_caps, webrtc_media_limitnot_negotiated_test, ignored_when_not_negotiated_testUnit tested
CCM-64.3.2, 4.3.4TSTR / TSTN, VBCMNot implemented (kept raw)

RFC 8285: header extensions

IDSectionRequirementModuleTestStatus
EXT-14.2One-byte form: profile 0xBEDE, ids 1..14, 1..16 data byteswebrtc_rtp_extone_byte_figure_test, encode_one_byte_testUnit tested
EXT-24.2Id 15 ends processing, its length ignored; earlier elements keptwebrtc_rtp_extid15_stops_testUnit tested
EXT-34.3Two-byte form: profile 0x100 + appbits, ids 1..255, 0..255 data bytes, appbits ignored on receivewebrtc_rtp_exttwobyte_figure_test, encode_forced_form_testUnit tested
EXT-44.1Zero padding bytes between elements and up to the 32-bit boundarywebrtc_rtp_extonebyte_figure_test, one_byte_errors_testUnit tested
EXT-54.2Id 0 is reserved for padding, not an elementwebrtc_rtp_extonebyte_errors_testUnit tested
EXT-66Mixed one-byte / two-byte use (extmap-allow-mixed): the encoder picks the two-byte form only when an element needs itwebrtc_rtp_extencodeauto_two_byte_test, prop_ext_roundtripUnit tested
EXT-74.1Malformed element lists rejected without crashingwebrtc_rtp_extevery_truncation_test, bad_rfc8285_ext_test, prop_rtp_mutation_never_crashesUnit tested
EXT-86extmap-allow-mixed answered only where offered, in effect when both descriptions carry itwebrtc_jsep_build, webrtc_jsep_paramswebrtc_media_ext_form_tests:negotiation_test, media_level_testUnit tested
EXT-94.1.2, 6Without it, one form throughout a stream: one-byte when every element we send fits, two-byte otherwisewebrtc_media_outone_byte_throughout_test, two_byte_throughout_test, long_mid_two_byte_testUnit tested
EXT-106With it, each packet takes the form its elements needwebrtc_media_outmixed_allowed_testUnit tested

RFC 4588: retransmission

IDSectionRequirementModuleTestStatus
RTX-14Payload is the OSN followed by the original payloadwebrtc_rtxrtx_testUnit tested
RTX-24Timestamp equal to the original timestampwebrtc_rtxrtx_testUnit tested
RTX-34SSRC multiplexing: distinct SSRC, own sequence numbers, RTX payload typewebrtc_rtxrtx_testUnit tested
RTX-44Packets without an OSN (padding only) are not retransmissionswebrtc_rtxrtx_padding_only_testUnit tested
RTX-55.3Receiver associates the RTX SSRC with the original stream (apt, rid / repaired-rid)webrtc_media, webrtc_media_ridwebrtc_media_tests:nack_rtx_test, webrtc_media_simulcast_tests:layer_nack_rtx_testUnit tested: ssrc-group:FID when announced, repaired-rid otherwise
RTX-66Sender keeps sent packets and avoids resending within one RTTwebrtc_nack_bufferbuffer_test, buffer_max_age_testUnit tested
RTX-7RFC 4585 §6.2.1, RFC 3711 §3.3NACK answered without RTX: the exact SRTP bytes of the first transmission are resent (no second protection of the same index)webrtc_media_outwebrtc_media_tests:nack_without_rtx_testUnit tested

RFC 2198 and RFC 8854: redundant audio and FEC

IDSpecRequirementModuleTestStatus
RED-12198 §3Block headers: F = 1, payload type, 14-bit timestamp offset, 10-bit length for redundant blocks; F = 0 and the payload type for the primary; blocks in header orderwebrtc_redlayout_test, two_blocks_test, primary_only_test, webrtc_red_gst_SUITEInteroperability tested (GStreamer rtpredenc and rtpreddec)
RED-22198 §3Blocks the header cannot describe (offset above 14 bits, length above 10 bits) are not sent; headers that do not match the payload are refusedwebrtc_redlimitstest, malformedtest, fuzz_testUnit tested
RED-32198 §3, §4, 8854 §7Receiver: the primary block is delivered as the packet it carries; a redundant block repairs the packet it repeats only when that packet was lost. A block carries no sequence number, so it is placed between the latest packet received before its timestamp and the RED packet (a sender may repeat an older packet than the previous one)webrtc_red, webrtc_media_inpackets_test, repair_distance_two_test, repair_without_duration_test, webrtc_pc_SUITE:red_repairs_loss, webrtc_red_gst_SUITE:gst_red_distance_1, gst_red_distance_2Interoperability tested (GStreamer, distance 1 and 2)
RED-48854 §4.2, 2198 §5audio/red offered after Opus with fmtp naming the primary payload types; answered with the offer's fmtp when all its payload types are keptwebrtc_jsep_capsofferedtest, answer_testInteroperability tested (Chromium answers it; Firefox, Pion and aiortc have no audio/red, see below)
RED-58854 §4.1, §8Sender: with red => true on a send, the previous payload goes with the new one; used on demand, not by default (FEC SHOULD follow the observed loss)webrtc_mediawebrtc_pc_SUITE:red_repairs_loss, webrtc_pc_interop_SUITE:erlang_sends_red_audio, webrtc_red_gst_SUITE:gst_decodes_our_redInteroperability tested (Chromium and GStreamer decode it)
RED-68854 §4.1RTP header parts not covered by redundancy (CSRCs, header extensions) approximated (RFC 2198 §4)Non-compliant / known limitation: a repaired packet keeps the RED packet's CSRCs and extensions
RED-78854 §5.1Video FEC: FlexFEC (RFC 8627) SHOULD be supportedwebrtc_flexfec, webrtc_media_out, webrtc_media_inFLEX-1 to FLEX-8Unit and E2E tested; no peer we test against implements RFC 8627 (libwebrtc has the draft-03 format behind a field trial): browsers answer without it and get ULPFEC (RED-9 to RED-14)
RED-88854 §4.1A send codec chosen by default is never a repair format (RED, RTX, ULPFEC, FlexFEC), whatever the offer's orderwebrtc_media_outdefault_send_codec_testUnit tested
RED-98854 §5.2, 5109 §7.3, §7.4ULPFEC packets: FEC header (E, L, P, X, CC, M, PT, SN base, TS and length recovery), one level 0 header with the protection length of the longest packet and a 16 or 48 bit maskwebrtc_ulpfecwebrtc_ulpfec_testsUnit tested
RED-105109 §8Recovery: an FEC packet with one of its packets missing gives it back (header bits, timestamp, length, payload), recovered packets help recover others; at most 160 packets and 32 FEC packets kept per streamwebrtc_ulpfecburstrecovery_test, fec_first_test, two_in_a_group_test, webrtc_ulpfec_gst_SUITEUnit and interoperability tested: GStreamer's rtpulpfecenc, every packet its masks leave recoverable rebuilt byte for byte
RED-118854 §5.2ULPFEC inside RED: media and FEC packets of a video stream in RED with one block, FEC computed over the media packet as it is without RED; video red, rtx for red and ulpfec offered (116, 117, 118) and answered, video RED without fmtp keptwebrtc_media_in, webrtc_media_out, webrtc_jsep_capswebrtc_media_fec_tests, webrtc_pc_interop_SUITE:erlang_sends_video_fecUnit and interoperability tested: Chromium and Firefox recover with our FEC packets (the last packet of every fifth frame dropped, no NACK: 60 of 60 frames decoded, against 20 without FEC)
RED-128854 §5.2Sender: FEC packets after each frame, interleaved over its packets (a burst of up to K losses repaired), at video_fec packets per media packet, or with auto twice the fraction lost the peer reports and at most one for two; none without losswebrtc_media_outrecovery_test, without_fec_testUnit tested
RED-133550 §6.4Recovered packets are delivered but not counted received: the loss reported keeps the sender protecting (as libwebrtc); FEC packets count as packets of the stream, are no gap for NACK or the frame bufferwebrtc_media_in, webrtc_frame_bufferrecovery_test, no_loss_testUnit tested
RED-14VP8 and VP9 frames with picture IDs are decodable by picture ID continuity, so a lost FEC packet breaks no frame chain (libwebrtc's reference finders)webrtc_frame_bufferrecovery_test, prop_webrtc_framesUnit and property tested
FLEX-18627 §4.2.1Repair packets on their own SSRC and sequence numbers, timestamp at sending time (90 kHz), M = 0, the protected SSRC in the CSRC listwebrtc_media_outwebrtcmedia_fec_tests:modes_test (flex)Unit tested
FLEX-28627 §4.2.2.1Flexible mask after SN base: 15, 46 or 110 bits, k = 1 when another mask block follows (as §4.2.2.1 writes it)webrtc_flexfecwebrtcflexfec_tests:burst_test, mask_sizes_testUnit tested
FLEX-38627 §4.2.2.2Fixed L columns and D rows read as rows (D <= 1) or columns (D > 1); L = D = 0 ignoredwebrtc_flexfecfixed_rows_and_columns_test, column_recovery_testUnit tested; only the flexible mask is sent
FLEX-48627 §4.2.2.3Retransmission repair packets: the source packet, R and F in place of the version, delivered when missingwebrtc_flexfec, webrtc_media_inretransmission_testUnit tested; none are sent (RTX repairs on NACK)
FLEX-58627 §6.2, §6.3Repair bit string of the first 16 header bits, the length past the fixed header, the timestamp and everything after the 12 byte header, shorter packets padded with zeros; recovery by XOR with the received packetswebrtc_flexfec, webrtc_ulpfecbursttest, modestestUnit tested
FLEX-68627 §5.1.2, §5.2.1video/flexfec at 90000 with repair-window: offered at 200 ms; answered only with a repair window of at most 1 s (160 packets kept), refused when absent or largerwebrtc_jsep_capswebrtcjsep_flexfec_tests:repair_window_testUnit tested; repair-window: with a colon (§7.1.2) is read too; the fmtp of the §7.1.1 example, a=fmtp:98; repair-window=..., lacks the space RFC 8866 §6.15 requires and is refused (rfc_example_test)
FLEX-78627 §7.1, RFC 5956 §4.3The repair SSRC offered and answered with a=ssrc-group:FEC-FR media repair, read from the remote description; repair packets also mapped by their CSRCwebrtc_jsep_build, webrtc_jsep_params, webrtc_media_inoffer_and_answer_testUnit tested
FLEX-88854 §5.1With FlexFEC and ULPFEC both negotiated, FlexFEC is sent (media not in RED)webrtc_media_outmodestestUnit tested; a repair packet protecting several SSRCs is read but not used for recovery, and no RTCP sender report is sent for the repair stream

Audio RED among WebRTC implementations (checked on 2026-11-17): libwebrtc implements it (Chromium). Firefox offers RED and ULPFEC for video only (JsepCodecDescription.h, media.navigator.video.red_ulpfec_enabled), Pion v4 has no RED payload type, and aiortc lists none. GStreamer's rtpredenc and rtpreddec are the independent implementation the format is checked against, over plain RTP.

RFC 8852 and RFC 8853: RTP stream identifiers and simulcast

IDSectionRequirementModuleTestStatus
RID-18851 4, 8852 3.1Every layer sent has its own SSRC and RTX SSRC and carries the MID and RtpStreamId header extensionswebrtc_mediawebrtc_media_simulcast_tests:send_layers_testUnit tested
RID-28851 4, 8852 3.2Retransmissions of a layer carry RepairedRtpStreamId, not RtpStreamIdwebrtc_mediawebrtc_media_simulcast_tests:layer_nack_rtx_testUnit tested
RID-38852 4, 8853 6.1RTCP SDES with the MID and RtpStreamId items for every sent layer, and a sender report per layerwebrtc_mediawebrtc_media_simulcast_tests:sdes_teaches_layers_testUnit tested
RID-49143 9.2, 8852 3.1Layers without a=ssrc lines demultiplexed by MID and rid; the SSRC is learned from the first packet, later packets may omit the extensionswebrtc_media_ridwebrtc_media_simulcast_tests:receive_layers_test, classify_testUnit tested, E2E (webrtc_pc_SUITE:simulcast_layers), interop (Chromium)
RID-58852 3.2RTX stream of a layer bound by repaired-rid; repair packets before their layer is known are droppedwebrtc_media_ridwebrtc_media_simulcast_tests:layer_nack_rtx_test, classify_testUnit tested
RID-68852 4SSRC to rid mapping learned from RTCP SDESwebrtc_media_ridwebrtc_media_simulcast_tests:sdes_test, sdes_teaches_layers_testUnit tested
RID-78853 5.2Streams with a rid that was not negotiated, or without any rid in a simulcast section, are not deliveredwebrtc_media_ridwebrtc_media_simulcast_tests:unknown_rid_dropped_test, classify_testUnit tested (counter unknown_rid)
RID-88853 6.1, 4585NACK, RTX and PLI per layer SSRC; keyframe request for one layer both wayswebrtc_mediawebrtc_media_simulcast_tests:layer_nack_rtx_test, layer_keyframe_testUnit tested, E2E, interop (Chromium reports the PLI on the requested layer only)
RID-98853 5.2, 5.3, RFC 7728Pause and resume of one layer; a layer the answer marks paused (~) starts paused and is resumed by RESUME once its SSRC is known from SDESwebrtc_media_pausewebrtc_media_simulcast_tests:layer_pause_test, answer_paused_layer_test, webrtc_pc_SUITE:simulcast_answererE2E tested

RFC 7728: RTP stream pause and resume

No independent implementation answers these messages (browsers, Pion and aiortc ignore them), so the rows stop at E2E tested.

IDSectionRequirementModuleTestStatus
PAUSE-18.1PAUSE-RESUME messages: RTPFB FMT 9, media SSRC 0, Target SSRC, Type, PauseID, PAUSED with the extended sequence number; reserved types ignored, unknown type specific data skippedwebrtc_rtcp_fbwebrtc_rtcp_fb_tests:pause_test, prop_webrtc_rtp:prop_rtcp_roundtripUnit tested
PAUSE-29a=rtcp-fb:* ccm pause offered with nowait for video; an offer is answered with the full implementation (config 1), nowait only when offeredwebrtc_jsep_caps, webrtc_jsep_paramswebrtc_jsep_caps_testsUnit tested
PAUSE-36.2, 8.1Sender: PAUSE with the current PauseID pauses (at once with nowait, after the hold-off otherwise; a RESUME during the hold-off cancels it), RESUME resumes and moves to the next PauseID, other PauseIDs are refused with REFUSED, a past RESUME while playing is ignoredwebrtc_rtp_pausewebrtc_rtp_pause_testsUnit tested
PAUSE-46.2PAUSED sent on pausing and repeated in the next two regular reportswebrtc_rtp_pause, webrtc_media_pausewebrtc_rtp_pause_tests:pause_resume_testUnit tested
PAUSE-56.1, 6.4Requester: PAUSE and RESUME retransmitted after 2 * RTT + T_dither_max (500 ms without an RTT) until PAUSED, REFUSED or, for RESUME, the stream's packets; the PauseID follows PAUSED and REFUSEDwebrtc_rtp_pausewebrtc_rtp_pause_tests:requester_testUnit tested
PAUSE-66.2A paused stream is not sent: send_rtp returns {error, paused} and the application is told (stream_paused, stream_resumed)webrtc_media_pausewebrtc_pc_SUITE:pause_resumeE2E tested
PAUSE-76.2Local pause by the sender, resume on BYE or time-out of the requester, TMMBR 0 as pauseNot implemented

RFC 5761 and RFC 8858: rtcp-mux

IDSectionRequirementModuleTestStatus
MUX-1RFC 5761 §4Demultiplex by the second octet: 192..223 is RTCPwebrtc_rtpclassifytestUnit tested
MUX-2RFC 5761 §4RTP payload types 64..95 MUST NOT be used when multiplexing: codecs on them are refused at new_peer_connection/1, never answered, and an answer using them is refused (payload_type_conflicts_with_rtcp)webrtc, webrtc_jsep_caps, webrtc_jsepwebrtc_jsep_tests:mux_payload_types_not_answered_test, mux_payload_type_answer_refused_test, mux_payload_types_refused_at_creation_testUnit tested
MUX-3RFC 8858 §4rtcp-mux-only: offers carry it with rtcp_mux_policy => require; an offer with it is answered with rtcp-mux (we always multiplex)webrtc_jsep_buildwebrtc_jsep_tests:rtcp_mux_only_testUnit tested

RFC 5506: reduced-size RTCP

IDSectionRequirementModuleTestStatus
RSIZE-13.4.2, 4.1Accept RTCP datagrams that do not start with SR / RRwebrtc_rtcpcompoundvalidation_testUnit tested
RSIZE-25Compound checks still available for peers that did not negotiate rtcp-rsizewebrtc_rtcpcompoundvalidation_testUnit tested
RSIZE-34.1, 5Regular reports are always compound; feedback alone only when rtcp-rsize is negotiated on every section, otherwise in a compound with an RR and the CNAMEwebrtc_media_outwebrtc_media_tests:compound_feedback_without_rsize_test, reduced_size_feedback_testUnit tested

Congestion control feedback

IDSectionRequirementModuleTestStatus
CC-1RFC 8888 §3.1Report blocks: SSRC, begin_seq, num_reports metric blocks of R, ECN, ATO; report timestampwebrtc_rtcp_ccfbwebrtc_rtcp_ccfb_tests:ccfb_test, ccfb_multi_and_empty_testUnit tested
CC-2RFC 8888 §3.1Odd number of metrics followed by 16 bits of zero paddingwebrtc_rtcp_ccfbccfb_testUnit tested
CC-3RFC 8888 §3.1R = 0: ECN and ATO ignoredwebrtc_rtcp_ccfbccfb_not_received_ignores_bits_testUnit tested
CC-4RFC 8888 §3.1At most 16384 metric blocks per report blockwebrtc_rtcp_ccfbccfblimits_testUnit tested
CC-5RFC 8888 §3.1ATO in 1/1024 s; 0x1FFE over range, 0x1FFF unavailablewebrtc_rtcp_ccfbatotestUnit tested
CC-6RFC 8888 §3.1num_reports counts metric blocks (range begin_seq to begin_seq + num_reports - 1, reported erratum)webrtc_rtcp_ccfbccfb_testUnit tested
TWCC-1draft-holmer-01 §2Transport-wide sequence number header extensionwebrtc_rtp_exttwcc_seq_testUnit tested
TWCC-2draft-holmer-01 §3.1Feedback header: base sequence, status count, 24-bit signed reference time (64 ms), feedback countwebrtc_rtcp_twccwebrtc_rtcp_twcc_tests:mixed_testUnit tested
TWCC-3draft-holmer-01 §3.1.3Run length chunks (draft example 1)webrtc_rtcp_twccrun_length_example_test, run_length_received_test, long_run_cut_testUnit tested
TWCC-4draft-holmer-01 §3.1.4One-bit and two-bit status vector chunks (draft example 1); chunks cut at the status countwebrtc_rtcp_twccone_bit_vector_example_test, mixed_test, status_count_cuts_vector_testUnit tested
TWCC-5draft-holmer-01 §3.1.1Symbol 11 is reserved and rejected (draft example 2 uses it)webrtc_rtcp_twccreserved_symbol_testUnit tested
TWCC-6draft-holmer-01 §3.1.5Receive deltas in 250 us: 8-bit unsigned small, 16-bit signed largewebrtc_rtcp_twccmixed_test, prop_twcc_roundtripUnit tested
TWCC-7draft-holmer-01 §3.1.5A delta that does not fit 16 bits starts a new feedback messagewebrtc_rtcp_twccfrom_arrivals_split_test, prop_twcc_arrivalsUnit tested
TWCC-8libwebrtcPadding with zero bytes or with the P bitwebrtc_rtcp_twccmixed_test, padding_bit_testUnit tested
CC-7RFC 8888 §3, §4, §6ack ccfb offered; when both sides have it and transport-cc is not negotiated, the receiver sends one CCFB report per 100 ms (per source the run from the first unreported sequence number to the highest, ECN and arrival offsets) and the sender turns the reports into estimator input, as transport-wide feedbackwebrtc_ccfb_receiver, webrtc_cc, webrtc_jsep_caps, webrtc_jsep_paramswebrtcccfb_tests, sim ccfb_converges_test, ccfbprobing_startup_test, webrtc_cc_SUITE:ccfb_without_transport_cc, ccfb_tracks_bottleneckE2E tested; transport-cc is preferred when both are negotiated (every browser sends it); probe clusters get their results from these reports too (sim ccfbprobing_startup_test: 1.4 Mbps in 200 ms against 4 s without probing)
TWCC-9draft-holmer-01 §2Every RTP packet of the transport (media and retransmissions, all transceivers) carries the next transport-wide sequence number when the peer accepted the extensionwebrtc_cc, webrtc_mediawebrtc_media_cc_tests:transport_wide_sequence_test, webrtc_cc_tests:twcc_seq_testUnit tested
TWCC-10draft-holmer-01 §3Receiver: feedback every 100 ms (twcc_interval) and on demand; numbers lost between two messages are reported; a late packet is reported in the next message with the range repeated from a bounded history; bounded pending arrivalswebrtc_twcc_receiverwebrtc_twcc_tests:interval_test, loss_between_feedbacks_test, late_packet_test, bounded_pending_test, gap_is_not_loss_test, prop_webrtc_cc:prop_twcc_feedback_roundtripUnit tested
TWCC-11draft-holmer-01 §3Sender: feedback mapped back to send time, arrival time and size; repeated reports ignored, lost then received reported again; bounded history (8192 packets, 10 s)webrtc_twcc_senderround_trip_test, wraparound_test, duplicate_feedback_test, unknown_and_old_test, age_limit_testUnit tested
TWCC-12draft-holmer-01 §3Feedback between two peers over UDP drives the sender's estimatewebrtc_media, webrtc_ccwebrtc_media_cc_tests:feedback_round_trip_test, webrtc_cc_SUITEE2E tested
REMB-1draft-alvestrand-rmcat-remb-03 §2.2PSFB FMT 15 "REMB": SSRC count, 6-bit exponent, 18-bit mantissa, SSRC listwebrtc_rtcp_fbremb_test, rr_remb_testUnit tested
REMB-2draft-alvestrand-rmcat-remb-03 §2A received REMB bounds the estimate; without transport-wide feedback it is the estimatewebrtc_gccwebrtc_gcc_tests:remb_test, webrtc_media_cc_tests:remb_testUnit tested
XR-1RFC 3611 §4.4Receiver Reference Time report blockwebrtc_rtcp_xrwebrtcrtcp_xr_tests:xr_test, xr_errors_testUnit tested
XR-2RFC 3611 §4.5DLRR report block with sub-blockswebrtc_rtcp_xrxrtest, xr_errors_testUnit tested
XR-3RFC 3611 §3Other block types kept with their type-specific bytewebrtc_rtcp_xrxr_unknown_block_testUnit tested
XR-4RFC 3611 §4.4A Receiver Reference Time block in each regular report while we receive streamswebrtc_media_outwebrtc_stats_SUITE:derived_membersE2E tested
XR-5RFC 3611 §4.5A DLRR block answering each Receiver Reference Time received (at most 16 sources), its delay in 1/65536 s; the answer to ours gives the round trip time of the streams we receivewebrtc_media_out, webrtc_media_inderived_membersE2E tested

Congestion control (RFC 8836, draft-ietf-rmcat-gcc-02)

The estimator is Google Congestion Control on the sender side, fed by transport-wide feedback (webrtc_gcc, webrtc_gcc_delay, webrtc_gcc_aimd). "sim" means the seeded bottleneck simulation of test/webrtc_cc_sim.erl and test/webrtc_cc_link.erl (one or two flows, 20 ms propagation delay each way, drop-tail queue of 300 ms), run by webrtc_gcc_tests.

IDSectionRequirementModuleTestStatus
GCC-1gcc-02 §5.2Pre-filtering: packets sent within burst_time (5 ms) form a group; a burst arriving after an outage is merged; out of order packets ignoredwebrtc_gcc_delaywebrtc_gcc_tests:reordering_and_burst_testUnit tested
GCC-2gcc-02 §5.3Arrival-time filter estimating m(i) from the inter-group delay variationwebrtc_gcc_delayconstant_delay_is_normal_test, growing_queue_is_overuse_test, draining_queue_is_underuse_testUnit tested: trendline estimator of libwebrtc (window 20, smoothing 0.9, gain 4), which the draft allows ("any adaptive filter"); the Kalman filter of the draft is not implemented
GCC-3gcc-02 §5.4Over-use after overuse_time_th (10 ms) above the adaptive threshold and m(i) not decreasing; under-use below minus the threshold; threshold not updated when the estimate exceeds it by more than 15, clamped to [6, 600], starting at 12.5webrtc_gcc_delaygrowing_queue_is_overuse_test, draining_queue_is_underuse_testUnit tested: K_u = 0.0087 and K_d = 0.039 from libwebrtc's trendline detector, not the draft values (0.01, 0.00018) tuned for the Kalman filter output
GCC-4gcc-02 §5.5Hold / Increase / Decrease transition tablewebrtc_gcc_aimdaimd_transitions_testUnit tested
GCC-5gcc-02 §5.5Multiplicative increase at most 8% per second far from convergence, additive (half a 1200-byte packet per response time of 300 ms for every flow; the draft takes the packet size from the flow's rate and the response time from its RTT, so a flow at a lower rate or on a longer path gained less and stayed below the others, while AIMD only converges to fairness when every flow adds the same amount (Chiu and Jain); 300 ms is the draft's RTT + 100 ms at libwebrtc's default RTT) near the average of previous decreases (EMA 0.95, three standard deviations, left once R_hat stays above it for a response time), A_hat < 1.5 R_hatwebrtc_gcc_aimdaimd_increase_test, webrtc_rmcat_tests (RMCAT-3)Unit tested; R_hat is the acknowledged rate over 500 ms of arrivals. As in libwebrtc, a decrease at an R_hat below the range also starts the average over, so a flow that fell with the capacity climbs back at the multiplicative rate (RMCAT-3)
GCC-6gcc-02 §5.5Decrease to beta R_hat, beta = 0.85webrtc_gcc_aimdaimd_decrease_testUnit tested; never above the current estimate, again only after one RTT unless R_hat fell below half the estimate (libwebrtc)
GCC-7gcc-02 §6Loss-based controller: below 2% x 1.05, 2% to 10% held, above 10% x (1 - 0.5 p); target = min(delay-based, loss-based)webrtc_gcclossdecrease_test, loss_decrease_negative_clock_test, moderate_loss_holds_test, loss_report_test, loss_report_then_feedback_test, sim random_loss_test, webrtc_cc_SUITEE2E tested; the ratio is taken over at least 20 packets or one second, every packet counts until a decrease and only packets sent after it count then, decreases at most once per 300 ms + RTT; RTCP receiver reports are used while no transport-wide feedback arrived, and the first transport-wide feedback releases the estimate they seeded
GCC-8gcc-02 §5.5Bounded memory, deterministic with injected timewebrtc_gcc, webrtc_ccsim deterministictestUnit tested
R8836-1RFC 8836 §2 (1)Low delay: queuing delay kept low once convergedwebrtc_gccsim converges500k_test, converges2m_test, converges10m_testSimulated: average target 0.95, 0.94 and 0.91 of the capacity at 500 kbps, 2 Mbps and 10 Mbps, average bottleneck queue 22 to 28 ms
R8836-2RFC 8836 §2 (1c, 1d)React quickly to a capacity reductionwebrtc_gccsim capacitydrop_test, webrtc_cc_SUITE:estimate_follows_capacity_dropSimulated and E2E tested: 2 Mbps to 500 kbps, the target is below 600 kbps within 2 s and the queue drains within 4 s
R8836-3RFC 8836 §2 (2)Fair to other instances of itselfwebrtc_gccsim fairnesstest, webrtc_rmcat_tests (RMCAT-4, RMCAT-5, RMCAT-8, RMCAT-10)Simulated over 24 seeds: three flows joining 20 s apart, after a pause and resume, and under ECN marking converge within 6 s of the last change and then keep a Jain index of 0.90 or more (0.934 to 1.000, 0.942 to 0.998 and 0.900 to 1.000); five flows with one-way delays of 10 to 150 ms reach 0.945 to 0.997
R8836-4RFC 8836 §2 (3)Do not starve competing TCP flows, avoid being starvedwebrtc_gccwebrtc_rmcat_tests (RMCAT-6, RMCAT-7)Simulated: against a long TCP flow the media keeps at least its minimum and TCP gets 1.5 to 1.7 Mbps of 2 Mbps; against ten short flows the media never drops below its minimum
R8836-5RFC 8836 §2 (4)Adapt quickly at the start of a flowwebrtc_cc_probe, webrtc_gcc_aimdsim probingstartup_test, probingslow_link_test, webrtc_cc_SUITE:probing_fills_link_at_startSimulated and E2E tested: probe clusters at 3 and 6 times the start bitrate, then at twice each good result, take a 2 Mbps link to 70% in 200 ms against 4 s without probing; on a 500 kbps link the result stays at what got through. Later probes (PROBE-6 to PROBE-8) adapt quickly again when the application was limited, after a drop and on a raised maximum
R8836-6RFC 8836 §2 (5)Stable when streams pause or are rate limitedwebrtc_gcc_aimdaimd_increase_testUnit tested: the estimate does not grow beyond 1.5 R_hat + 10 kbps
R8836-7RFC 8836 §2 (6)One estimate for all streams of the transportwebrtc_ccwebrtc_media_cc_tests:transport_wide_sequence_testImplemented: one estimate per BUNDLE transport; the split between streams is left to the application
R8836-8RFC 8836 §2 (8)Feedback over RTCP with AVPFwebrtc_mediawebrtc_media_cc_tests:feedback_round_trip_testUnit tested
R8836-9RFC 8836 §2 (9)Use DSCP markingswebrtc_dscp, webrtc_media_outsee TR-20 to TR-28 of transport.mdImplemented: media is marked by priority (RFC 8837); the estimator treats every mark alike
R8836-10RFC 8836 §2 (10)React to a lack of feedbackwebrtc_gccfeedback_timeout_testUnit tested: 20% cut per feedback_timeout (1.5 s) while sending without feedback
R8836-11RFC 8836 §2 (11)Stable with AQMwebrtc_gccwebrtc_rmcat_tests (RMCAT-10)Simulated with CoDel marking (RFC 8289, 5 ms target): one flow follows the capacity at 0.95 of its share, three flows share the link with a Jain index of 0.99, queue p95 under 50 ms. Dropping AQMs are not simulated
NADA-1RFC 8698NADA congestion controlNot implemented

RFC 8867: evaluation test cases

These are the RMCAT test cases, run on the simulator by webrtc_rmcat_tests with the settings of RFC 8867 §4:

  • 50 ms one-way delay, and packet delay variation as RFC 8868 §4.5 asks (NR-BPDV, 5 ms standard deviation, cut at 3, no reordering within a flow);
  • for the ECN cases, CoDel marking at its default 5 ms target and 100 ms interval (RFC 8289);
  • a 300 ms tail-drop queue;
  • video from 150 kbps to 1.5 Mbps starting at 150 kbps, with 20 kbps audio on the same transport;
  • TCP from webrtc_cc_tcp: RFC 5681 congestion control with SACK recovery (RFC 6675).

Every case runs on six seeds by default and on 24 with RMCAT_SEEDS=24 (the seed sets link jitter and flow variation), and senders take turns at the bottleneck in a new order every millisecond: in a fixed order the last one always queued behind the others and got a weaker delay signal. The bounds are fixed from each case's expected behavior:

  • after a change, a flow averages 0.7 to 1.05 of its share within 10 s;
  • flows that share the link converge (RFC 8867 §4.1) within 40 s of the last join or resume: from then on, over the rest of the run, their Jain index stays at 0.9 or more;
  • 5.3 is compared with its reference case, the same run with an unimpaired backward link (§5.3): the reference meets every bound, and the impaired run keeps its share, with a queue above 100 ms only by as much as its feedback waits on the backward link;
  • the 95th percentile of the queue stays under 100 ms;
  • media never drops below its minimum;
  • TCP keeps a quarter of the link.
IDSectionCaseTestStatus
RMCAT-15.1Variable capacity (1, 2.5, 0.6, 1 Mbps), one flow, 50 and 100 msvariablecapacity_single_50ms_test, variablecapacity_single_100ms_testSimulated over 24 seeds: 0.84 to 1.00 of the available rate in every phase, queue p95 at most 98 ms
RMCAT-25.2Variable capacity (4, 2, 3.5, 1, 2 Mbps), two flowsvariablecapacity_multiple_testSimulated over 24 seeds: together 0.85 to 1.00 of the available rate, queue p95 at most 73 ms
RMCAT-35.3Media both ways, forward and backward capacities changingfeedbackcongestion_testSimulated over 24 seeds: reference run 0.87 to 1.00 of its share, queue p95 at most 104 ms (one phase of seed 20 above the bound, see PERFORMANCE.md); with the backward link congested the forward flow keeps 0.80 to 1.00 (queue p95 at most 99 ms), the backward flow 0.90 to 1.00
RMCAT-45.4Three flows joining 20 s apart on 3.5 Mbpscompetingflows_testSimulated over 24 seeds: converged at the third join, then Jain index 0.934 to 1.000, 0.93 to 0.95 of the link, queue p95 at most 80 ms
RMCAT-55.5Five flows with one-way delays of 10 to 150 ms on 4 Mbpsrttfairness_testSimulated over 24 seeds: converged once the last flow joined, then Jain index 0.945 to 0.997, 0.94 to 0.95 of the link, queue p95 at most 53 ms. GCC's additive increase per RTT + 100 ms gave 0.79 to 0.91: the increase no longer depends on the RTT or the rate (GCC-5)
RMCAT-65.6One flow against a long TCP flow, 300 and 1000 ms queueslongtcp_300ms_test, longtcp_1000ms_testSimulated over 24 seeds: media 193 to 242 and 318 to 474 kbps, never below its minimum; TCP 1672 to 1718 and 1447 to 1584 kbps
RMCAT-75.7Two flows against ten short TCP flowsshorttcp_testSimulated over 24 seeds: media 330 to 481 kbps, never below its minimum, Jain index between them above 0.9; TCP 274 to 325 kbps
RMCAT-85.8Three flows, one paused from 40 to 60 spauseresume_testSimulated over 24 seeds: the others take the paused share; converged at the resume, then Jain index 0.942 to 0.998
RMCAT-96.1Flows with prioritiesNot implemented: needs coupled congestion control (RFC 8699)
RMCAT-106.25.1 and 5.4 with RFC 8888 feedback and ECN markingecnvariable_capacity_test, ecncompeting_flows_testSimulated over 24 seeds: one flow 0.82 to 1.00 of the available rate, queue p95 at most 58 ms; three flows converge within 6 s of the third join, then Jain index 0.900 to 1.000
RMCAT-116.3Multiple bottlenecksNot tested: the simulator has one bottleneck per direction

Bandwidth probing

IDSourceRequirementModuleTestStatus
PROBE-1libwebrtc ProbeControllerWith transport-wide feedback, the first video packet starts clusters at 3 and 6 times the start bitrate, each at least 15 ms at its rate and 5 packets, paced at its ratewebrtc_cc_probewebrtc_cc_probe_tests:initial_clusters_testUnit tested
PROBE-2libwebrtc ProbeBitrateEstimatorResult from the cluster's feedback: at least 80% received, send and receive intervals within 1 s, the smaller of the send and receive rates, 95% of the receive rate when below 90% of the send ratewebrtc_cc_probeestimate_test, lost_probe_test, no_feedback_testUnit tested
PROBE-3libwebrtc ProbeControllerA result of at least 70% of the probed rate probes again at twice it, up to max_bitratewebrtc_cc_probeestimate_testUnit tested
PROBE-4libwebrtc DelayBasedBwe, SendSideBandwidthEstimationThe delay-based estimate takes a probe result unless over-use is detected; the loss-based estimate is lifted to it too (a lower max_bitrate or a feedback timeout may have cut it)webrtc_gccsim probingstartup_test, webrtc_cc_SUITE:probing_on_raised_maximumE2E tested
PROBE-5RFC 3550 §5.1, RFC 4588Probes are padding-only packets with transport-wide sequence numbers, on the RTX stream when negotiated, else on the media stream; received padding-only packets are not delivered to the applicationwebrtc_media_out, webrtc_media_inwebrtc_cc_SUITE:probing_fills_link_at_startE2E tested
PROBE-6libwebrtc AlrDetector, ProbeControllerApplication limited: media (padding left out) under 65% of the estimate for 500 ms, until back over 80%; meanwhile a cluster at twice the estimate every 5 swebrtc_cc_probe, webrtc_ccwebrtccc_probe_tests:alr_probe_test, sim probing_application_limited_testSimulated: a flow sending 400 kbps on a 3 Mbps link sits at 1.5 times what is acked, then the probes take the estimate to the link
PROBE-7libwebrtc ProbeController (RequestProbe)After a drop under 66% of the previous estimate, while application limited or within 3 s of it, leaving over-use within 5 s probes at 85% of the rate before the drop, when that would still raise the estimate, at most once every 5 s; a sender that fills its share does not probe after a drop (the drop came from competing traffic or a smaller link, and probing refills the queue)webrtc_cc_probedropprobe_test, no_drop_probe_when_not_app_limited_test, sim probing_after_drop_testSimulated
PROBE-8libwebrtc ProbeControllerA raised max_bitrate (set_congestion_control/2) probes the new maximum when the estimate sat at the old onewebrtc_cc_probe, webrtc_ccmax_raised_test, webrtc_cc_SUITE:probing_on_raised_maximumE2E tested
PROBE-9libwebrtc ProbeController, PacingControllerOne cluster in flight; probes after the start limited to 10 a minute; media leaving during a cluster is part of it (tagged in the transport-wide history), padding only makes up the difference with the cluster's rate, a cluster is complete with its bytes and 5 packetswebrtc_cc_probe, webrtc_ccboundstest, media_in_cluster_test, sim probing_with_media_in_flight_testSimulated: after a dip on a 3 Mbps link still carrying 700 kbps of video the probe measures 2.4 to 2.8 Mbps, against 1.3 to 1.5 Mbps when only padding was counted

Pacing

IDSectionRequirementModuleTestStatus
PACE-1libwebrtc, RFC 8836 §2 (1f)Leaky bucket at pacing_factor (2.5) x target, burst bounded to burst_ms (10 ms, at least one packet); never more than rate x interval + burst + one packetwebrtc_pacerwebrtc_pacer_tests:timing_accuracy_test, next_timeout_test, idle_burst_is_bounded_test, prop_webrtc_cc:prop_pacer_conservationUnit tested
PACE-2libwebrtcAudio before retransmissions before video, each in orderwebrtc_pacerpriority_test, prop_pacer_accountingUnit tested
PACE-3Bounded queue (1000 packets): the oldest video packet is dropped first, drops counted per class; no packet lost or duplicated otherwisewebrtc_pacerbounded_queue_drops_oldest_video_test, bounded_queue_without_video_test, prop_pacer_accountingUnit tested
PACE-4libwebrtcQueuing delay bounded by max_queue_ms (1000 ms) by sending fasterwebrtc_pacerqueue_delay_bound_testUnit tested
PACE-5Pacing applies once feedback backs the estimate; it can be turned off at run time (webrtc:set_congestion_control/2)webrtc_ccwebrtc_cc_tests:unpaced_without_feedback_test, paced_with_estimate_test, pacing_off_test, webrtc_media_cc_tests:pacing_test, webrtc_cc_SUITE:pacing_can_be_disabledE2E tested
PACE-6libwebrtcRetransmissions limited to the target bitrate over one secondwebrtc_ccrtx_limit_testUnit tested

RFC 8083: RTP circuit breakers

IDSectionRequirementModuleTestStatus
CB-14.1RTCP timeout: sending without any report on the stream for 3 x Td, Td with Tmin = 5 s (15 s)webrtc_circuit_breakerwebrtc_circuit_breaker_tests:rtcp_timeout_test, reports_prevent_rtcp_timeout_test, idle_stream_test, webrtc_media_cc_tests:rtcp_timeout_test, reports_keep_breaker_quiet_testUnit tested
CB-24.2Media timeout: MEDIA_TIMEOUT consecutive reports with a non-increasing extended highest sequence number while sending, k = 5webrtc_circuit_breakermedia_timeout_test, media_progress_resets_testUnit tested; Tf (frame interval) is not known to this layer and is left out of MEDIA_TIMEOUT
CB-34.3Congestion: sending rate above 10 x the simplified TCP throughput over CB_INTERVAL reports, loss fraction weighted by the reporting intervalswebrtc_circuit_breakercongestion_test, congestion_needs_rtt_testUnit tested; the 10 G Tf term of CB_INTERVAL is left out
CB-44.4ECN circuit breaker: with RFC 8888 feedback, RTP goes out ECT(0) where the socket can set and read marks (RFC 6679); marks reported cleared stop it, ECT or CE validate it; CE marked packets count as lost in the next report block the circuit breakers see and in the loss-based estimate (RFC 3168 §5)webrtc_cc, webrtc_transport_net, webrtc_circuit_breakerwebrtcccfb_tests:ecn_validation_test, ce_counts_as_loss_test, sim ce_marking_bottleneck_test (RED-like marking: 0.94 of capacity, 4 ms of queue), webrtc_cc_SUITE:ccfb_without_transport_cc, ccfb_ce_marking (a CE marking bottleneck)E2E tested
CB-54.5Cease transmission when the RTCP or media timeout breaker triggers; no automatic restart; restart rate limitedwebrtc_cc, webrtc_mediawebrtc_cc_tests:circuit_breaker_halts_test, webrtc_cc_SUITE:circuit_breaker_halts_streamE2E tested: RTP of the transceiver stops (send_rtp returns {error, {circuit_breaker, Reason}}, queued packets are dropped), RTCP continues; webrtc:resume_stream/2 is refused for circuit_breaker_hold (15 s)
CB-64.3Congestion breaker: the sender MAY first cut its rate by ten, and MUST cease if it triggers again over the next CB_INTERVALwebrtc_ccwebrtc_cc_tests:congestion_breaker_two_steps_test, webrtc_cc_SUITE:estimate_follows_capacity_dropUnit tested: first trigger caps the target at a tenth of the sending rate and re-arms, second trigger halts; the cap is lifted after the hold time without a new trigger
CB-7RFC 8834 §7.1WebRTC endpoints implement the circuit breaker; with peers that only send receiver reports the sender stays within its limitswebrtc_cc, webrtc_gccwebrtc_gcc_tests:rr_only_peer_testUnit tested: receiver reports drive the loss-based estimate, seeded with the sending rate, and pacing applies

Payload formats

IDSectionRequirementModuleTestStatus
VP8-1RFC 7741 §4.2Payload descriptor: X, N, S, PID, I (7 or 15-bit picture ID), L, T, Kwebrtc_rtp_vp8webrtcrtp_vp8_tests:vp8_parse_test, vp8_minimal_test, vp8_truncated_testUnit tested
VP8-2RFC 7741 §4.3Key frame: S = 1, PID = 0, P bit of the payload header 0webrtc_rtp_vp8vp8_parse_test, vp8_minimal_testUnit tested
VP8-3RFC 7741 §4.2Picture ID and TL0PICIDX rewritten keeping the field widthwebrtc_rtp_vp8vp8_rewrite_testUnit tested
VP8-4RFC 7741 §4.4Frame split into payloads with S set on the firstwebrtc_rtp_vp8vp8_packetize_testUnit tested
H264-1RFC 6184 §5.6Single NAL unit packetswebrtc_rtp_h264webrtc_rtp_h264_tests:single_nal_testUnit tested
H264-2RFC 6184 §5.7.1STAP-A: F is the OR, NRI the maximum of the aggregated unitswebrtc_rtp_h264stap_a_testUnit tested
H264-3RFC 6184 §5.8FU-A fragmentation and reassembly; S and E never both setwebrtc_rtp_h264fua_test, h264_errors_testUnit tested
H264-4RFC 6184 §5.8Fragments of a NAL unit with a lost fragment are discardedwebrtc_rtp_h264fu_a_testUnit tested
H264-5RFC 6184 §5.7.1, §6.4Interleaved mode: STAP-B with its DON, a unit alone in a STAP-B since single NAL unit packets and STAP-A are not allowedwebrtc_rtp_h264stapb_test, interleaved_errors_testUnit tested
H264-7RFC 6184 §5.7.2MTAP16 and MTAP24: DONB the first unit in decoding order, DOND and timestamp offset per unitwebrtc_rtp_h264mtap_testUnit tested
H264-8RFC 6184 §5.8FU-B for the first fragment with the DON, FU-A for the others; S and E never both setwebrtc_rtp_h264fu_b_test, deinterleave_fragments_testUnit tested
H264-10RFC 6184 §8.2.2packetization-mode 2 negotiated as its own codec when the application configures it (codecs option), never against mode 1webrtc_jsep_capswebrtc_jsep_caps_tests:h264_interleaved_testUnit tested
H264-9RFC 6184 §5.5, §7.2De-interleaving: units leave in DON order once sprop-interleaving-depth + 1 VCL units wait, the rest on flushwebrtc_rtp_h264deinterleave_test, prop_webrtc_rtp:prop_h264_interleaved_roundtripUnit tested; no peer we test against sends mode 2, and the buffer ignores sprop-deint-buf-req, sprop-init-buf-time and sprop-max-don-diff
H264-6RFC 6184 §5.3Key frame detection by IDR (type 5), including STAP-A and FU-A startwebrtc_rtp_h264stap_a_test, fu_a_testUnit tested
OPUS-1RFC 7587 §4.2One Opus packet per RTP payload, never fragmentedwebrtc_rtp_opuswebrtcrtp_opus_tests:opus_testUnit tested
OPUS-2RFC 7587 §4.1RTP clock rate 48000webrtc_rtp_opusopustestUnit tested
OPUS-3RFC 6716 §3.1, §3.2Packet duration from the TOC byte and frame count, at most 120 mswebrtc_rtp_opusopustestUnit tested
VP9-1RFC 9628 §4.2Key frame: P = 0, B = 1, spatial layer 0webrtc_rtp_vp9webrtcrtp_vp9_tests:vp9_testUnit tested
AV1-1AV1 RTP v1.0.0 §4.4Aggregation header; key frame by N = 1webrtc_rtp_av1webrtcrtp_av1_tests:av1_testUnit tested
VP8-5RFC 7741 §4.4, §4.5Depacketization: descriptors removed, the VP8 data of the frame's packets joinedwebrtc_frame_codecwebrtcframe_buffer_tests:in_order_test, webrtc_frames_gst_SUITE, webrtc_pc_interop_SUITE:peer_sends_framesUnit and interoperability tested: GStreamer's rtpvp8pay, Chromium and Firefox, decoded by ffmpeg
VP8-6RFC 6386 §9.1Width and height of a key frame: frame tag, start code 9d 01 2a, 14 bit sizeswebrtc_rtp_vp8framesize_test, no_size_in_delta_testUnit and interoperability tested
VP8-7RFC 7741 §4.2Packetizer sets the temporal layer index and layer sync bit (T, TID, Y)webrtc_rtp_vp8temporal_layers_testUnit tested
VP9-2RFC 9628 §4.2, §4.2.1Descriptor: picture ID, layer indices with TL0PICIDX in non-flexible mode, up to three P_DIFF in flexible mode, scalability structure (layer sizes, picture group)webrtc_rtp_vp9vp9_descriptor_test, webrtc_corpus_testsUnit tested
VP9-3RFC 9628 §4.3Packetizer: B on the first and E on the last payload of a layer frame, scalability structure in the firstwebrtc_rtp_vp9vp9descriptor_test, in_order_testUnit tested
VP9-4RFC 9628 §4.4Depacketization: the layer frames of a picture (one timestamp, marker on the last) joinedwebrtc_frame_codecinorder_test, webrtc_frames_gst_SUITE, peer_sends_framesUnit and interoperability tested: rtpvp9pay, Chromium and Firefox
VP9-5VP9 bitstream §6.2Key frame size from the uncompressed header (frame marker, profile, sync code, color config), or the highest layer of the scalability structurewebrtc_rtp_vp9framesize_testUnit and interoperability tested
H264-11RFC 6184 §8.4, ITU-T H.264 §7.4.1.2.3Depacketization to Annex B; an access unit starts with an AUD, SPS, SEI or the slice of macroblock 0, taken as the first packet only once the packets before it are given up; the last SPS and PPS before the frame, in sequence order, put before an IDR that lacks them; an IDR without them is not decodablewebrtc_frame_codech264_parameter_sets_test, prop_webrtc_frames:prop_frames_survive_network, webrtc_frames_gst_SUITE, peer_sends_framesUnit, property and interoperability tested: rtph264pay with config-interval=-1, Chromium; libwebrtc's SPS/PPS tracker refuses such an IDR too
H264-12ITU-T H.264 §7.3.2.1.1, §7.4.2.1.1Picture size from the SPS: exp-Golomb fields, high profile chroma and scaling lists, croppingwebrtc_rtp_h264framesize_test (174 x 130, cropped)Unit and interoperability tested
H264-13ITU-T H.264 §7.4.2.1.1, Annex A.3.1Every SPS field checked against its range as it is read (id 31, bit depths 6, log2_max_frame_num 12, POC type 2, POC cycle 255, reference frames 16, picture side 2048 macroblocks); no count from the stream sizes an allocationwebrtc_rtp_h264sps_poc_cycle_bound_test, sps_field_ranges_test, prop_webrtc_parsers:prop_h264_sps_fields, prop_decoders_never_raise (heap limited)Unit and property tested: a 13 byte SPS with a 2^31 frame cycle no longer exhausts memory
AV1-2AV1 RTP v1.0.0 §4.4, §4.5Depacketization: OBU elements with or without length fields (W), fragments continued with Z and Y, a temporal unit in the low overhead format: a temporal delimiter, then every OBU with its size field (AV1 §5.2, §7.5)webrtc_rtp_av1av1_fragments_test, av1_obu_count_test, webrtc_corpus_tests (libwebrtc's assemble-frame corpus), webrtc_frames_gst_SUITE, peer_sends_framesUnit and interoperability tested: rtpav1pay, Chromium and Firefox, decoded by ffmpeg
AV1-3AV1 RTP v1.0.0 §5Packetizer: temporal delimiters and tile lists left out, OBUs without size fields, every element with a length (W = 0), N on the first payload of a coded video sequencewebrtc_rtp_av1av1fragments_test, in_order_testUnit tested
AV1-4AV1 §5.5Maximum frame size from the sequence header (timing, decoder model and operating points skipped)webrtc_rtp_av1framesize_testUnit and interoperability tested

Frame assembly and jitter buffer

With frames on a transceiver, webrtc_frame_buffer turns the packets of each received stream into frames.

IDSectionRequirementModuleTestStatus
FRAME-1RFC 3550 §A.1Packets in extended sequence order, duplicates dropped, packets of a frame already gone discarded (packets_discarded)webrtc_frame_bufferreorderand_duplicates_test, late_packet_test, prop_webrtc_framesUnit and property tested
FRAME-2RFC 3550 §5.1A frame is the packets of one timestamp, complete from its first packet (marked by the format, or after a packet of another frame) to the one with the marker bitwebrtc_frame_bufferinorder_test, prop_frames_survive_networkUnit and property tested; timestamps must rise with sequence numbers, so H.264 with B frames (not used by WebRTC encoders) is not supported
FRAME-3RFC 7741 §4.2, RFC 9628 §4.2A frame leaves once decodable: a key frame; VP8 and non-flexible VP9 base frames after the previous base frame (TL0PICIDX), higher layer frames after their base frame and every frame between, or when layer sync; VP9 flexible frames once every referenced picture left; other video in sequence after a key frame; audio alwayswebrtc_frame_buffertemporallayers_test, vp9_flexible_test, loss_breaks_chain_test, joinlate_test, audio_testUnit tested
FRAME-4RFC 3550 §6.4.1Playout time: timestamp plus the smallest transit of the last 10 s plus the target delay, three times the interarrival jitter of frames within min_delay and max_delaywebrtc_frame_bufferplayout_time_testUnit tested; libwebrtc also weighs frame sizes, not done here
FRAME-5playout-delay extensionThe sender's minimum and maximum delay bound the target delay when negotiatedwebrtc_frame_buffer, webrtc_media_inplayout_delay_extension_testUnit tested
FRAME-6RFC 4585 §6.2.1A frame missing packets waits for a later frame's playout time plus 100 ms, or three round trips when NACK repairs the stream, or until NACK gives up on a missing packet; then it and the frames that need it are dropped (frames_dropped)webrtc_frame_bufferretransmissiontest, lostreport_test, loss_breaks_chain_testUnit tested and interoperability tested against GStreamer with loss and reordering
FRAME-7RFC 4585 §6.3.1, RFC 5104 §4.3.1After a drop, a key frame request: PLI when negotiated, otherwise FIR with a new sequence number for each request; at most one per round trip and five a second. webrtc:request_keyframe/2 and NACK use the same choicewebrtc_media_inwebrtc_media_frames_tests:pli_on_loss_test, fir_only_test, application_request_fir_testUnit tested
FRAME-8Bounded: 2048 packets and 3 s per streamwebrtc_frame_bufferbound_test, prop_garbage_never_crashesUnit and property tested
FRAME-9Frames of independent packetizers decode: GStreamer (VP8, VP9, H.264, AV1, in order and with one packet in 40 lost and reordering), Chromium (VP8, VP9, H.264, AV1) and Firefox (VP8, VP9, AV1), with as many frames and key frames as the browser encodedwebrtc_frame_bufferwebrtc_frames_gst_SUITE, webrtc_pc_interop_SUITE:peer_sends_framesInteroperability tested; ffmpeg decodes what GStreamer sent without an error, at the encoded size and with the same key frames

Known limitations

  • Congestion control clocks are integer milliseconds: arrival times in transport-wide feedback are multiples of 1 ms although the format carries 250 us.
  • webrtc_rtp:encode/1 does not check that the payload type avoids 64..95: the peer connection refuses such codecs (MUX-2), so only a caller encoding packets itself can produce one.
  • webrtc_nack (receiver NACK generation) follows RFC 4585 §6.2.1 for the message and libwebrtc for the retry policy (retry every max(RTT, min interval), give up after max_retries or max_age, request a key frame when more than max_missing packets are missing); RFC 4585 does not specify this policy.