Module 1 — When Voice Became Data

Module 1 Examination: When Voice Became Data

You have 75 minutes. The examination has three parts: Part A is code, graded per check with partial credit and hints disabled; Part B is forensic diagnosis of canned artifacts; Part C is concepts. The pass mark is 80 percent. Everything runs offline in this page — no camera, no microphone, no network.

ACoding: the wire, byte by byte

Since RFC 5761 (April 2010, Perkins and Westerlund), RTP and RTCP typically share a single port, and every receiver's first job is telling them apart: read the second byte of the packet as a full 8-bit value — 192–223 means RTCP (Sender Report = 200, Receiver Report = 201), while an RTP packet with a dynamic payload type of 96–127 puts 96–127 there (marker bit clear) or 224–255 (marker bit set), never 192–223. The capture in Problem A1, supplied as CAPTURE (an array of Uint8Arrays), is such a mixed stream: Opus on payload type 111 at the 48 kHz RTP clock (RFC 7587), interleaved with the sender's SR and the far end's RR. All multi-byte fields are big-endian.

Part A — Coding tasks≈ 50% of the examination

Three problems: demultiplex and parse a mixed RTP/RTCP capture, reconstruct stream statistics across a sequence-number wrap, and build the μ-law quantization pipeline.

BForensics: four incidents

Each scenario below presents a canned artifact — counters shaped like getStats output, a spectrogram, packet timelines. Read the artifact, then answer the corresponding question in the quiz that follows. One question per scenario.

Incident 1 — choppy but lossless. A receiver reports steady packetsReceived, packetsLost = 0, and strictly contiguous sequence numbers — yet the audio is choppy, with audible gaps and stutters. The jitter buffer is repeatedly starving and refilling.

Incident 2 — the robotic tone. A test harness plays a pure 5 kHz tone into a legacy 8 kHz G.711 gateway whose anti-aliasing filter has been misconfigured out of the path. Listeners describe the result as "robotic." The spectrogram of what the far end hears is sketched in Figure E1.1: the sent tone never appears; a strong component sits at 3 kHz instead.

Incident 3 — the timestamp cliff. An RTP timeline from a 48 kHz Opus stream shows this around the twelve-second mark:

seq=1999   ts=159040    arr=11.980 s
seq=2000   ts=160000    arr=12.000 s
seq=2001   ts=400000    arr=17.023 s      ts jumped +240,000 ticks = 5.0 s
seq=2002   ts=400960    arr=17.043 s

Exhibit B.3 — Incident 3 artifact: sequence numbers contiguous, RTP timestamp discontinuous by five seconds of media time.

Incident 4 — one identity, two voices. A capture at the receiver shows packets bearing a single SSRC arriving from two different transport addresses, with two interleaved sequence-number and timestamp series:

ssrc=0x5A13F00D  pt=111  seq=4021   ts=1728000    from 203.0.113.7:49170
ssrc=0x5A13F00D  pt=111  seq=17755  ts=88970240   from 198.51.100.9:61044
ssrc=0x5A13F00D  pt=111  seq=4022   ts=1728960    from 203.0.113.7:49170
ssrc=0x5A13F00D  pt=111  seq=17756  ts=88971200   from 198.51.100.9:61044

Exhibit B.4 — Incident 4 artifact: the same SSRC from two sources, two independent sequence/timestamp series.

Part B — Forensic diagnostics≈ 30% of the examination

One question per incident. Diagnose before you look at the options.

CConcepts: the numbers that hold it up

Part C examines the quantitative bedrock of the module — Nyquist (1928) and Shannon (1949) on sampling, the ITU-T G.114 guidance that one-way mouth-to-ear delay should stay at or below 150 ms, the split of labor between sequence numbers and timestamps, and RFC 3550's rule that RTCP traffic is held to 5 percent of session bandwidth.

Part C — Concept questions≈ 20% of the examination

Five questions. Do the arithmetic; the wrong options are the standard mistakes.