Upsampling with the Cambridge Audio 840C: measurements

[ddaudio]

I've been experimenting with upsampling, using a Cambridge Audio 840C, which has digital inputs and outputs as well as a CD drive and allows upsampling of digital input to digital output.

The capture chain for these tests is LMS (Lyrion Music Server - ex Logitech) --> Squeezelite --> USB to spdif (SMSL PO100) --> Cambridge Audio 840C --> spdif to minicomputer with on-board toslink input, captured via ALSA/SoX.

Test data are 1khz sine waves files generated by REW at 16/44.1 and 24/96. Graphs are 32k FFT, Hann window, using REW.

First of all, check the transparency of the capture chain, with the 840C set to Pass Through:

16/44.1 looks fine (local file direct green, captured red):

As does 24/96 (direct blue, captured red):

First question: why the bigger "spread" of the 1 khz tone towards lower levels? Is this just an artifact of the measurement settings?


Now, upsampling with the 840C from 16/44.1 (red) to 24/48 (blueish) and 24/96 (green):
(all upsampling with the dither set to on in the 840C menu, but I tried with it off and there was no appreciable difference)

So leaving aside the open question above about the "spread" at higher sample rates, the main difference seems to be a reduction in the peak level by about 2dB when upsampling. Other than that, at least with the 840C, upsampling to 24/48 seems technically superior to 24/96.

My first thought is that upsampling seems pointless. Any comments, suggestions for improving the test, etc (or an answer to the "spread" issue)?

As an aside, the 840C upsamples everything to 24/384 for its analogue outputs. This is not configurable.

The blurb from the 840C manual:

The 840C incorporates a raft of new technologies and features. Key to
its abilities is the ATF (Adaptive Time Filtering) upsampling process -
developed in conjunction with Anagram Technologies of Switzerland.
This system intelligently interpolates 16 bit/44.1kHz CD (or other) data
to 24bit/384kHz through the use of a 32 bit Analog Devices Black Fin
DSP (Digital Signal Processor) for the very best sound quality. The ATF
system applies sophisticated polynomial curve fitting interpolation and
incorporates a time domain model which allows data buffering and re-
clocking almost completely eradicating digital jitter.
Because the audio data rate is so high aliasing artefacts are moved way
above audible frequencies allowing us to use a low order 2 pole linear-
phase Bessel filter on the output for constant group delay and minimal
phase shift.
Two very high quality Analog Devices AD1955 24 bit DACs (Digital to
Analog Converters) are used in dual differential mode.

 

[staticV3]

First question: why the bigger "spread" of the 1 khz tone towards lower levels? Is this just an artifact of the measurement settings?

Same FFT size with higher sample rate means larger FFT bins means larger curtain around the fundamental.

Increase the FFT size in line with the sample rate and you keep bin size equal, resulting in an equally sized curtain.

Or use Rectangular window with periodic matched signals to get rid of this curtain effect altogether.

 

[ddaudio]

Same FFT size with higher sample rate means larger FFT bins means larger curtain around the fundamental.

Increase the FFT size in line with the sample rate and you keep bin size equal, resulting in an equally sized curtain.

Or use Rectangular window with periodic matched signals to get rid of this curtain effect altogether.

Thanks. I obviously have a lot to learn!

I'm not yet sure about how to implement the second approach (rectangular window). Presumably I'd need to capture exactly the same length from the soundcard as the original, which is hard to do.

On changing the FFT size, it moves the apparent noise floor (which I understand is not real): does that make visual comparisons invalid? Conversely, is it valid to draw conclusions from the comparison 44.1/48/96 above (all with a 32khz FFT size)?