Asynchronous Sample Rate conversion (ASRC) DACs

[jkeny]

What are people's view on ASRCs? I find that an ASRC benefits highish to medium jitter sources but invariably masks the benefits of low jitter sources.
I always am suspicious when a number of different inputs into a DAC all sound the same! It could be that the DAC is doing such a good job of jitter reduction that it makes all inputs behave at their best but in my experience, with DACs that have defeatable ASRCs, they sound better without ASRC when using low jitter sources. Even the ESS DACs, that have a supposedly very sophisticated ASRC, sound better when ASRC is turned off.

Any experiences to relate?

 

[DonH50]

<Off-topic as I have no direct experience to relate. It is an interesting subject so hopefully this post will not be offensive.>

Obviously many things other than jitter can cause sonic degradation, and it is difficult to isolate the effects of ASRC from everything else going on in the design, including the effect of sample rate conversion. Having all sources sound the same, I am not sure exactly what that means. If jitter is eliminated as an error source, then all sources should sound as good as the low-jitter sources, yes? Any other effects, assuming all esle is the same, must be related to the sample rate conversion. SRC itself introduces all sorts of other variables, natch.

Random thoughts - Don

 

[jkeny]

<Off-topic as I have no direct experience to relate. It is an interesting subject so hopefully this post will not be offensive.>

Obviously many things other than jitter can cause sonic degradation, and it is difficult to isolate the effects of ASRC from everything else going on in the design, including the effect of sample rate conversion.

Sure, it's difficult to isolate the effect of ASRC & implementation is obviously important as are many other technical factors. Of course, I'm generalising in my comments but they are based on a lot of users experiences.

Having all sources sound the same, I am not sure exactly what that means.

What I mean is Toslink, SPDIF & USB sound the same fed into the DAC - no discernible difference between these sources when ASRC is on but when off (some DACs can turn it on/off), the low jitter sources sound noticeably better than they do when ASRC is on & the sources now sound different from one another. Conclusion, ASRC is bringing all inputs to the same level, improving some sources but dis-improving others

If jitter is eliminated - Any other effects, assuming all esle is the same, must be related to the sample rate conversion. SRC itself introduces all sorts of other variables, natch.

Random thoughts - Don

 

[jkeny]

Let's look a bit closer at what ASRC does - it takes an input data rate & converts it to a completely different output data rate so we immediately see that the fundamental job of the Asynchronous Sample Rate Converter is to provide output audio samples corresponding to points of time IN BETWEEN the original audio samples. Sometimes only one sample in between, sometimes two ... because the output rate is faster. It's job is therefore one of interpolation or filling in the samples between the original samples. This interpolation is fraught with issues. This interpolation introduces error because it can never be perfect - the higher the ASRC bandwith & the higher the number of interpolation points the lower the error. This is probably one of the reasons that the high sampling rates of the ESS DACs & recently introduced NAD both sound better than most ASRCs but neither are perfect.

Here's an explanation of how ASRC works in an ESS DAC -taken from here http://www.anedio.com/index.php/article/multi_stage_jitter_reduction

The final refinement is performed by the ES9018 DAC, with its unique sample-rate conversion technology (patented by ESS Technology, US patent 7330138). It transports the original samples to a completely new clock domain, run by an ultra low phase noise oscillator. The new clock domain is at a much higher frequency (80 MHz in the case of the D1 DAC). Initially, the audio data is oversampled to the new domain simply by duplicating the samples, as indicated by the repeated dots in the diagram. Because the two sample frequencies are unrelated, the sample times of the two clocks will not align exactly. Note that almost all the samples of the output are simple replicas of the input. Only at the point where the input sample changes does the timing mismatch become an issue.

As shown in the figure zoomed into a transition point, the output sample clock falls slightly after the input sample clock, which causes a portion of the signal to be missing (the shaded area). It is this missing area that must be corrected. The key to ESS Technology's innovation lies in transforming this sampling-time mismatch into a single intermediate sample that corrects for the timing error. At the transition point, it generates a new intermediate sample in such a way to match the shaded area caused by the timing mismatch. Thus, the time error is transformed into an equivalent digital level.
To calculate the shaded area, the position of the input clock relative to the reference clock must be determined accurately. This critical task is performed by a DPLL (Digital Phase-Locked Loop), which locks onto the input clock and compares it to the precision low phase noise reference. The whole process is extremely accurate, achieving errors less than -175 dB.

Now, even though the statement is made that an error of < -175dB is achieved, the ESS DAC still sounds different between sources with different jitter levels.

 

[Andre Marc]

What are people's view on ASRCs? I find that an ASRC benefits highish to medium jitter sources but invariably masks the benefits of low jitter sources.
I always am suspicious when a number of different inputs into a DAC all sound the same! It could be that the DAC is doing such a good job of jitter reduction that it makes all inputs behave at their best but in my experience, with DACs that have defeatable ASRCs, they sound better without ASRC when using low jitter sources. Even the ESS DACs, that have a supposedly very sophisticated ASRC, sound better when ASRC is turned off.

Any experiences to relate?

My Bryston BDA-1 sounds better with async upsampling selected. It converts 44.1 multiples to 176.4 and 96 to 192.

 

[jkeny]

My Bryston BDA-1 sounds better with async upsampling selected. It converts 44.1 multiples to 176.4 and 96 to 192.

AFAIK, that is Synchronous Sample Rate Conversion (SSRC) not Asynchronous (ASRC).
The difference between the two types is large & fundamental.
SSRC uses an integer multiplier to convert from one clock to another because these clocks are integer multiples of one another i.e 44.1 is multiplied by 4 to get 176.4. The conversion of data in this scenario doesn't require any complex interpolation or guessing of the data as is required in ASRC where the new clock is a non-integer derivation of the original clock.

Even so, does a low jitter 16/44 source sound better when upsampled to 176.4 than when played back natively? Does 16/176.4 (upsampled) sound as natural as the 16/44.1? I reckon it probably will & will benefit from the higher sample rate easing of the digital interpolation filter. Where I find that there is a problem is in using ASRC which seems to sound unnatural.

 

[jkeny]

As further grist to the mill - here's a site that compares various SRC software upsamplers. A "correct" SRC algorithm should have a "perfect" passband, and "infinite" attenuation past the passband. In addition, there should be no additional harmonic distortion generated. The criteria is simple, but very difficult to achieve. As can be seen very few SRC algorithms perform well. These examples are of software implementations.

Using the correct algorithm & achieving the accuracy & computational power required in a small piece of silicon is a mammoth task. So hardware implementation of SRC is very precarious, in my opinion.

 

[Vincent Kars]

ASRC is indeed not trivial: http://www.iet.ntnu.no/courses/fe8114/slides/upsanddownsofasrc.pdf
But even straight up or down sampling e.g. from 44 to 176 and back is not easy because of the filtering.
Uli Bruggeman has a nice example: http://www.aktives-hoeren.de/viewtopic.php?f=30&t=2242&start=44

http://thewelltemperedcomputer.com/KB/SRC.htm

 

[jkeny]

Yes, Vincent, good links.
Partly this thread started as a result of Vincent trying my USB to SPDIF device & finding little difference between it & a TOSLINK feed into his Benchmark DAC.

My experience so far from 100s of customers is that ASRC is ALWAYS detrimental to the sound when using a low jitter source such as my USB-SPDIF device. It has benefits for high/med jitter sources but masks the benefits achievable from low jitter sources. In other words it converges all sources to a middle ground sound.

Vincent, is your experience with your Benchmark DAC that it's compulsory ASRC is always beneficial to the sound? It's difficult to evaluate this as you can't turn it off!

 

[Vincent Kars]

In itself it is an interesting hypothesis that there is a trade off somewhere when using ASRC.
Best would be somebody with a AP kit measuring the jitter performance using the various inputs.

The same happened to me when listening to the VLink.
A marked difference between the adaptive mode USB of the Benchmark and the async Vlink over SPDIF or AES/EBU.
I didn’t hear any difference between SPDIF and AES/EBU.
But there are people who claim to hear a world of differences.
Sometimes I had the feeling there was a difference between the Toslink and the SPDIF but I really needed a true sighted test to hear this

It might be a limitation in my hearing and/or perception but this applies to all of us.

Anyway I’m happy with the Benchmark as it offers the functionality I need (various digital inputs, pre-amp, headphone) and it delivers a transparent sound.

I do think it is nice to know that if my next computer comes without Toslink I can easily obtain the same sound quality by using an async USB to SPDIF converter.

One day I might decide to buy a pre-amp with a async USB input, that will probably be the best solution as no outside clock is involved.

 

[Andre Marc]

AFAIK, that is Synchronous Sample Rate Conversion (SSRC) not Asynchronous (ASRC).
The difference between the two types is large & fundamental.
SSRC uses an integer multiplier to convert from one clock to another because these clocks are integer multiples of one another i.e 44.1 is multiplied by 4 to get 176.4. The conversion of data in this scenario doesn't require any complex interpolation or guessing of the data as is required in ASRC where the new clock is a non-integer derivation of the original clock.

Even so, does a low jitter 16/44 source sound better when upsampled to 176.4 than when played back natively? Does 16/176.4 (upsampled) sound as natural as the 16/44.1? I reckon it probably will & will benefit from the higher sample rate easing of the digital interpolation filter. Where I find that there is a problem is in using ASRC which seems to sound unnatural.

Forgive me John, yes you are 100% correct. I reversed the question in my head.

The BDA-1 does indeed use SSRC.

I DO find that 44.1 converted to 176.4/24 does sound more natural.

What is interesting is that for several years I was a HUGE upsampling skeptic. I really thought it was another audiophile buzzword.
After hearing what it can do, and understanding the basic theory behind it, I am a convert.

I am fan of DACs that offer defeatable upsampling.

 

[DonH50]

I think a lot of the improvement has to do with reducing the impact of the image filter in the audio band, which upsampling allows one to do. That is, upsampling itself may not do much (though if done wrongly can mess things up), but using a lower-order filter at a much higher frequency at the DAC's output vs. the filter required at 44.1 kS/s can have audible benefits.

IMO - Don

 

[jkeny]

I think a lot of the improvement has to do with reducing the impact of the image filter in the audio band, which upsampling allows one to do. That is, upsampling itself may not do much (though if done wrongly can mess things up), but using a lower-order filter at a much higher frequency at the DAC's output vs. the filter required at 44.1 kS/s can have audible benefits.

IMO - Don

Yes, I concur - the easing of the conditions for the digital interpolation filter is to my mind where the sonic advantages with SSRC come from (no or simple data sample interpolation needed).
The issue with ASRC seems to be the accuracy needed for the complex data sample interpolation required & having to achieve this in real time. Most implementations of ASRC seem to get this wrong & the advantages of the filter easing just don't compensate for the errors/distortions introduced in the interpolation stage.

The ESS DACs go some way towards achieving this but the implementation details are scarce. The NAD Master series seems to take a further step in upsampling to 844KHz & using 35bit arithmetic. In the case of the ESS DACs using synchronous clock still sounds better than asynch clocks. The NAD has no way to bypass the ASRC. In both the ESS & NAD different input sources still sound different.

 

[DonH50]

I was actually talking about the analog output filter, that is the image filter after the DAC chip, but of course you are right that the digital input filter (especially for a delta-sigma DAC) is also in play.

 

[amirm]

The issue with ASRC seems to be the accuracy needed for the complex data sample interpolation required & having to achieve this in real time. Most implementations of ASRC seem to get this wrong & the advantages of the filter easing just don't compensate for the errors/distortions introduced in the interpolation stage.

That's a tough argument to make since companies show that the noise floor of their interpolator is below that of the source. I think a better hypothesis is that asynchronous sample rate converters have the same problem PLLs have in that they still have to track the source clock. If they do not, then their interpolation ratio is wrong. If the thing jumps back and forth to stay in sync, such error would not show up as jitter since output clock is fixed. Yet it will create non-linear distortion. I have yet to see anyone analyze this problem even though it is a known issue here.

 

[edorr]

FWIW, the PS Audio Perfectwave DAC MKII has optional sample rate conversion (no conversion = native mode) and optional asynchronous mode (= nativeX mode). You can do asynchronous only in the native sample rate. In my experience NativeX consistently sounds the best. This conclusions holds even when the source has a very good clock (I have an Offramp 5 USB converter with dual turboclock upgrade).

 

[LL21]

FWIW, the PS Audio Perfectwave DAC MKII has optional sample rate conversion (no conversion = native mode) and optional asynchronous mode (= nativeX mode). You can do asynchronous only in the native sample rate. In my experience NativeX consistently sounds the best. This conclusions holds even when the source has a very good clock (I have an Offramp 5 USB converter with dual turboclock upgrade).

Edorr - Just so i (a non-techie) understand, are you saying you prefer to listen to D/A done in the original native rate of how the recording was made? So no upsampling, oversampling? Is this different from those 'old school' Non-Oversampling DACs? If so, how? Thanks for any help!

 

[jkeny]

No, I think he is saying that he prefers ASRC (NativeX mode).
edorr, in what ways does it sound better to you?

 

[edorr]

No, I think he is saying that he prefers ASRC (NativeX mode).
edorr, in what ways does it sound better to you?

You are both correct. NativeX is both the native sampling rate (no upsampling) AND it is asynchronous. I am not good at describing differences in sound quality, but recently I found that what I have been getting through all my upgrades is less "grain". Music is just becoming more fluid, with less "edge".

 

[jkeny]

You are both correct. NativeX is both the native sampling rate (no upsampling) AND it is asynchronous. I am not good at describing differences in sound quality, but recently I found that what I have been getting through all my upgrades is less "grain". Music is just becoming more fluid, with less "edge".

Now I'm confused - ASRC means that the sampling rate will be changed to a new sampling rate, usually a higher sampling rate? Are you talking about something else?

Edit: I just looked up the Pefectwave & see where you are mistaken - the NativeX mode is asynchronous USB mode, not Asynchronous Sample Rate Conversion (ASRC) - different beast altogether