Thorsten and gang have been a little quiet of late. We gave them a nudge and here is something they asked us to share with you.
Measurements Matter (Not!) Part I of III – Hearing a fly fart at 20 feet
Noise. Signal-to-Noise Ratio (SNR)
Our team has just spent some time running the complete and working prototype (after redesigning the PSU and that little hiccup with the J-Fet switching) of the iDSD micro through the Audio Precision 2.
The test bench.
The micro iDSD has actually turned in a more than respectable performance.
I hasten to add that up to now there is little proven link between audio measurement results and subjectively perceived sound quality. Most standardised measurements originated in advertising and were standardised to avoid excessive misuse (but not completely eradicated ).
Many a designer who relies on measurements only, finds himself bothered, bewildered and thoroughly confused when many a listener prefers a better sounding product that measured distinctly poorer than his own.
That said, as some have already asked about measurements and it has never hurt to have respectable measured results, as long as the subjective listening results are also great
Well, here goes, the iDSD micro measured performance data. Please keep all these caveats in mind. The executive summary of the test results reads:
Line Out Level: 2.15V
SNR: 117dB A-weighted
Line THD+N: < 0.003% (@0dBFS)
HP THD+N: < 0.008% HP (@0dBFS and 0.5W/16Ohm simulated headphone load)
Jitter: below AP System 2 measurement limit
The signal-noise ratio may seem "only very good" but what must be remembered is that we are running the line out at 2V (nominal, industry-standard) level and as usual, the analogue stage is the noise limit, not the DAC.
We could have boosted the SNR figure to make it look really sexy by choosing to set a higher line out level:
- for example setting the line out level to 4.25V would have produced a 6dB increase in SNR 117dB un-weighted and 123dB A-weighted*, which reads great but in the real world, this would have resulted in much less usable volume control range for anything we are driving. Even as it stands we are only 3dB (0.5 Bit) off a 20 Bit equivalent SNR**.
* The use of “A-weighting” is a long standing standard for dynamic range and SNR measurements in Digital Converters (ADC/DAC). All Datasheets nowadays quote the “A-weighted” number. Thus for consistency we always include A-weighted and unweighted SNR/DNR.
** This is the true measure of any DAC’s or ADC’s resolution, also called ENOB (Equivalent Number Of Bit’s), for example a certain DAC that is promoted as “32 Bit/384KHz” actually shows a SNR of 100dB which is actually 16.5 Bit ENOB, despite all that 32 Bit stuff, so in analogue terms it has around halve the bits claimed…
For anyone who likes pretty graphs, we have oodles, this is the first of several select ones with short comments:
Graph 01 - iDSD micro White Noise Line 100k 1X 2X 4X 8X
This shows the frequency response of the iDSD micro using noise loading with different sample rates
192K (purple) and
plus the system noise floor at 384kHz.
- As the sample rate goes up we can see that the filtering of high frequencies is relaxed and by the time we are at 192kHz and above the filtering is very gentle, maximising time-domain (impulse) response fidelity.
- We could show a square wave, but it would look essentially perfect, so little point.
- The slight trade-off price is a little higher leakage of the ultrasonic images of the (noise) signal, however as most of it is 120dB down on full scale this is not very worrisome. I remember a professional recording engineer once describe -120dB as “Fly farts at 20 feet”…
We also take requests (for measurements) just ask.
(to be cont’d.) Part II: Distortion
Edited by iFi audio - 6/6/14 at 1:56am