Clocking
Packet time is not conversion time.
The receiver’s audio clock determines FIFO reads and DAC cadence. Ethernet timing is recovered independently for transport. A buffer helps manage packet variation; it does not erase a permanent rate mismatch.
Three operating cases
Rate control depends on the source.
Receiver-paced elastic source
The receiver grants a cumulative group limit. The source can pace production in response, while outstanding data in flight remain inside the grant. Duplicate grants do not add credit; an older grant does not retract the current one.
Common-frequency live source
When source and receiver share a disciplined reference, the design may keep long-term rates related. This must be measured as bounded FIFO occupancy and relative drift. A shared 10 MHz reference establishes frequency discipline, not a shared sample epoch.
Independent fixed-rate live source
A producer whose clock cannot be controlled eventually fills or drains any finite buffer when its rate differs from the receiver. Credits cannot change that physical clock. The system must provide valid conversion, a declared bound, or safe rejection and mute behaviour.
A finite margin
Buffering delays a mismatch. It does not remove it.
For relative frequency error ε, available margin M sample groups, and group rate r, approximate time to a buffer limit is M / (r × |ε|). A 1 ms margin at 50 ppm lasts about 20 seconds without correction.
Packet phase, network delay, packet reserve, source response time, processing stalls and in-flight frames affect first-failure timing. The 20-second value is analytical and does not predict a hardware measurement.
The reference tier shall provide a specified external 10 MHz input, independent audio clock generation, lock monitoring and controlled holdover. Entry-level operation can use an internal oscillator with receiver feedback. Invalid timing or expired safe holdover causes local mute or abort. A STATUS message does not delay protection or trigger automatic restart.
For raw DSD, an all-zero bitstream can create a large DC condition. Use converter mute or a validated DSD idle/transition sequence; do not substitute zeros as a generic mute.
Measurement
Claims require clock and occupancy evidence.
The draft calls for output frequency and uncertainty, relative drift, phase-noise spectra with stated integration limits, discrete spurs and time-domain period or phase error. Testing must cover both frequency families, local and external-reference modes, idle and admitted traffic, temperature changes, reference interruption and reacquisition.
Log FIFO occupancy and cumulative group counts during the same tests. A synthesiser’s lock signal is not a substitute for showing bounded occupancy, safe holdover or upstream rate control. None of those prototype measurements is claimed as complete.