I made it to about page 7 of this very interesting thread before giving up on trying to find the answer: why do those industrial SAR chips have such excellent THD and poor S/N or SINAD?
The switched capacitor DAC surely is akin to an R2R in that it does not have perfect linearity, so I would expect an AD that uses a DAC with a nonlinearity to behave just like a classic R2R DAC, i.e. good noise performance and poor distortion. However, the data sheets (such as for LTC2378-20) are just the opposite. Why?
The switched capacitor DAC surely is akin to an R2R in that it does not have perfect linearity, so I would expect an AD that uses a DAC with a nonlinearity to behave just like a classic R2R DAC, i.e. good noise performance and poor distortion. However, the data sheets (such as for LTC2378-20) are just the opposite. Why?
^^^
I'll give you one good reason for sigma delta ADCs: the audio grade sigma delta ADCs (with the notable exception of a now defunct AKM part) use noise shaping to "clean" the noise at audio frequencies, and push it at HF (over 50KHz). This greatly improves the SNR up to 96KHz sampling rate, at the expense of the SNR at sampling rates of 192KHz or higher.
Why are they doing this, since those noise levels, with or without noise shaping, are not audible, anyway? I would suspect it's simply for the "look at my SNR" bragging rights, and because it essentially costs nothing to do.
The LTC2378-20 is a 20bit SAR ADC, to the extend I am aware of there are no audio grade SAR ADCs, so you cannot compare apple and pears.
I'll give you one good reason for sigma delta ADCs: the audio grade sigma delta ADCs (with the notable exception of a now defunct AKM part) use noise shaping to "clean" the noise at audio frequencies, and push it at HF (over 50KHz). This greatly improves the SNR up to 96KHz sampling rate, at the expense of the SNR at sampling rates of 192KHz or higher.
Why are they doing this, since those noise levels, with or without noise shaping, are not audible, anyway? I would suspect it's simply for the "look at my SNR" bragging rights, and because it essentially costs nothing to do.
The LTC2378-20 is a 20bit SAR ADC, to the extend I am aware of there are no audio grade SAR ADCs, so you cannot compare apple and pears.
Last edited:
Yes, I know all that. But to my understanding, the switched capacitor DAC has to rely on geometric ratios between the elements and you can never get that completly right in lithography. So it is inherently non-linear, much like an R2R DAC is. So how does an SAR ADC that really consists of a non-linear DAC and a comparator get this kind of THD performance? And why is its noise so poor, where it should really shine?
I think if a 2 kHz full scale signal is used, it is almost an apples to apples comparison. Oh wait, there's the wider bandwith, that might explain the poorer SN (at least if they don't limit that in the analysis).
Let's look at page 6 of the LTC2378-20 data sheet
https://www.analog.com/media/en/technical-documentation/data-sheets/237820fb.pdf
The first graph in the middle row is an 128 k FFT of a 2 kHz full scale signal. It would do any sigma delta ADC proud. There are no harmonics nor spuriae down to at least -145 dB. I am aware of no audio ADC that can do this. And this with an inherently non-linear DAC? Or is it that the bins are so fine that you can't see the harmonics?
The last picture in that row is what I really don't get. Why would THD (not THD+N) degrade with signal frequency, going from -135 dB near 0 Hz to -113 dB at 25 kHz? Is that the switched capacitor DAC becoming more nonlinear because of not enough time for parasitic capacitances in the switches to discharge?
Anyway, if your aim is to look at harmonics of a 20 kHz signal, I don't really see that this ADC has any advantages over current top of the line audio sigma-delta ADCs. They are definitely better on the harmonics, and averaging will take care of the noise.
I think if a 2 kHz full scale signal is used, it is almost an apples to apples comparison. Oh wait, there's the wider bandwith, that might explain the poorer SN (at least if they don't limit that in the analysis).
Let's look at page 6 of the LTC2378-20 data sheet
https://www.analog.com/media/en/technical-documentation/data-sheets/237820fb.pdf
The first graph in the middle row is an 128 k FFT of a 2 kHz full scale signal. It would do any sigma delta ADC proud. There are no harmonics nor spuriae down to at least -145 dB. I am aware of no audio ADC that can do this. And this with an inherently non-linear DAC? Or is it that the bins are so fine that you can't see the harmonics?
The last picture in that row is what I really don't get. Why would THD (not THD+N) degrade with signal frequency, going from -135 dB near 0 Hz to -113 dB at 25 kHz? Is that the switched capacitor DAC becoming more nonlinear because of not enough time for parasitic capacitances in the switches to discharge?
Anyway, if your aim is to look at harmonics of a 20 kHz signal, I don't really see that this ADC has any advantages over current top of the line audio sigma-delta ADCs. They are definitely better on the harmonics, and averaging will take care of the noise.
I don't know that the noise is "poor". It's probably as syn08 said, the audio parts use aggressive noise shaping. The other part you have to consider is bandwidth. A lot of datasheets use A-weighted or band-limited numbers for SNR. If you look at LTC2387-18, which is a part I've used before in a non-audio application, it would reach respectable noise figures if you filtered and decimated to audio bandwidths/rates.
I lack detailed knowledge of the internals, but I wonder if there aren't compromises in some of these parts to preserve DC specs and INL/DNL that are not present in audio converters.
I lack detailed knowledge of the internals, but I wonder if there aren't compromises in some of these parts to preserve DC specs and INL/DNL that are not present in audio converters.
Last edited:
Nothing is perfect in the real world, but on the other side nothing, in any present of near future IC process, is better defined than the geometric ratios by lithography.
I also think you are in a pretty common, perhaps counter intuitive, pitfall. Averaging does not lower noise. Synchronous averaging does, which is not something that can be done with a common sound card and PC software (or at least I have not seen it). And of course, decreasing the bin size lowers the noise floor, otherwise said, it increases the process gain.
Try to compare the THD+N @10KHz input with your favorite sound card, and a 20bit SAR, say at 192KHz sample rate, and you will be surprised on the result (or not). Delta sigma ADCs don't have an ENOB over 20bit (according to my measurements, the ADS127L01, which is a great part, has about 19.6) so there's no reason to believe they can get a better SNR without noise shaping.
Last edited:
Look, while I am interested in this ESS ADC & spectrum plots, as the DAC uses hyperstream modulation TOO, NOT SDM !!!
The result within the ESS DAC's are lower noise floor...
Hp
Just watch the videos given Two Old ESS Talks | Audio Science Review (ASR) Forum with some details about...
As I watched the presentation regarding hyperstream modulation...
. 1/2 gain of input as SDM NEEDS, is not required
. they do not have any explanations why this multi mixing in math terms really work, but performs at the end. So you will see lower noise floor
. so hyperstreaming is different from the ordinary SDM as various feedback models used & SDM orders. So it is keep as secret
I'd argue this is absolutely still SDM, just an optimization. I would not believe everything you hear in a marketing pitch.
Also, keep in mind the implementation is not the same for an ADC as a DAC. In a DAC the modulator is entirely digital. In an ADC, there have to be analog components. Who knows if these techniques are even applicable or if the modulator has any relation to what is in their DACs.
How is it that AKM and Cirrus have no trouble matching the ESS noise performance if they don't have "hyperstream"...
Also, keep in mind the implementation is not the same for an ADC as a DAC. In a DAC the modulator is entirely digital. In an ADC, there have to be analog components. Who knows if these techniques are even applicable or if the modulator has any relation to what is in their DACs.
How is it that AKM and Cirrus have no trouble matching the ESS noise performance if they don't have "hyperstream"...
Bill,
Some discussion at: ESS Sabre Reference DAC (8-channel) ...Post #2643
Its a thing modulators do. Not sure about how that would work in an ADC. Might have to give it some DC input and see what happens...
Some discussion at: ESS Sabre Reference DAC (8-channel) ...Post #2643
Its a thing modulators do. Not sure about how that would work in an ADC. Might have to give it some DC input and see what happens...
If an audio ADC has DC at it's input surely someone has got the design wrong OR I the test set has been specified to handle a certain offset with acceptance of any performance changes?
Its a measurement technique only. Makes it easier to measure a performance metric. Modulator guys know about it and test that way, at least they do for dacs. Regarding DC, audio dacs aren't useful at DC either, but if you send them a DC signal in PCM format you can map out noise floor verses signal amplitude.
Last edited:
I said noise, I meant noise. In the general sense, across the product line where there are comparable parts. I don’t really care about your latest obsession 100+ dB down. No offense, of course.
Btw what is funny is that if you look at the second graph that you linked to, the CS4398 actually has the flattest DNR vs code / DC offset of all DACs tested including the Sabres.
Last edited: