Higher Sampling Rate -> Higher THD+N!

[Flecko]

I have found something interesting for all people not using the NOS1 or an upsampling DAC. The picture below shows a graph that shows distortion of the PCM1794A for different sample rates over room temperature. The interesting thing is, that with higher sampling rate, the distortion increases! This is not what most would expect. I looked in the specs of the PCM1704 and it also produces higher distortion with higher sampling frequency. This may explain my prefered setting of 44Khz. There are for sure advantages of higher sampling rates but I gues if you like to benefit from it, the DAC must be designed to this frequency [Example: A possibility would be to use a "softer" filter with higher sample frequency. So less phase shift is produced]. And one can imagine, if your dac will produce higher distortion with higher sample rate, your cable and your usb device/soundcard might do this too! Higher frequencies means higer energies and more stress to the components. So if you got a normal OS dac, a soundcard or something else that is not the NOS1 or an upsampling dac, it can be better to use the native sampling frequency, because that is what your dac will be tuned to (worst case cenario). [And also ArcPred should not be used, it creates additional distortion if you have no NOS (Peter please corect me if I am wrong but I think you wrote this somewhere else).]

[PeterSt]

Hi there Adrian,

I looked in the specs of the PCM1704 and it also produces higher distortion with higher sampling frequency.

I am not aware of this, nor do I measure it ! So, where is this to be found ? (page of the datasheet)

Further ... it would be a kind of logical I think; The more the chip has to process within a time unit, the harder it gets. But you just said that yourself.

Another thing maybe - personally I would not know how to interpret (or test) a chip to these kind of merits. And then I mean, test it at different sample rates for THD(+N). This is because the lower sample rate just bears inherently more distortion. But I don't know everything, so maybe never mind that one.

May it help : I measure not any difference between 192 and 384, while my analyser won't sample beyond 192. So, it keeps on sampling at 192 never mind the resolution is 384 in the mean time, and I'd think if this were true for the NOS1, figures would drop.
And also : maybe when measuring 48 output (or 44.1) figures would be relatively better, but this is masked by the low output sample rate. Well, you have seen the plots in the other topic I think; there's no escape from that, i.e. the chip possibly performing better at that rate is masked all over with the "bad" output. Still it would make sense.

Peter

[Flecko]

I am not aware of this, nor do I measure it ! So, where is this to be found ? (page of the datasheet)

Sorry, was a mistake. I misinterpreted the table on the first page. Nevertheless, the behavior of the PCM1704 could be the same as the PCM1794A.

Another thing maybe - personally I would not know how to interpret (or test) a chip to these kind of merits. And then I mean, test it at different sample rates for THD(+N). This is because the lower sample rate just bears inherently more distortion. But I don't know everything, so maybe never mind that one.

Measuring conditions were the same for all three sample rates according to the PCM1794A datasheet. THD+N is proportional to the sample rate, (table of page 3). 

May it help : I measure not any difference between 192 and 384, while my analyser won't sample beyond 192. So, it keeps on sampling at 192 never mind the resolution is 384 in the mean time, and I'd think if this were true for the NOS1, figures would drop.

In the case of PCM1794A, there is always 8x oversampling used. What bb has measured is for systemclock>352. But THD+N could go down for even lower systemclock!? I don't know how to measure THD+N but shouldn't it be possible to meassure it for a certain frequency range (20-20k Hz), while sample rates are different (like they did in the bb sheet)? Thinking about it, it seems strange if there were the absolute same THD+N for different systemclocks.

And also : maybe when measuring 48 output (or 44.1) figures would be relatively better, but this is masked by the low output sample rate. Well, you have seen the plots in the other topic I think; there's no escape from that, i.e. the chip possibly performing better at that rate is masked all over with the "bad" output. Still it would make sense.

I think in the case of NOS1, it should be good with high fs, because the "upsampling" is part of the filter. It could get worse if the filter is designed for 44.1, and you just double to 88.2 without consider that in the filter. But I don't know how the doubling is done. I am speculating very muche here.

...time to sleep...