DSP Measurements and Rising Noise Floor

[John_Siau]

I imagine most DACs are perfectly fine if they have digital volume control. Has anyone seen evidence to the contrary?

Michael

Here is the evidence to the contrary:

There are several ASRC chips that include an AES/SPDIF receiver but lack a digital volume control. When these ASRC chips are used in a DAC, they will precede the digital volume control and the DAC will clip intersample overs at all volume settings.

The ESS DAC chips have AES/SPDIF receivers built in, and the chip includes a digital volume control. Unfortunately the ESS chip will clip intersample overs even when the volume is turned down. Many ESS DACs will have a problem.

Given these two factors, I would expect to see plenty of DACs that have intersample clipping problems at all digital volume settings.

In many cases, you have to add additional chips to provide a digital volume control in the beginning of the signal path. The "cookbook" data sheet circuits often don't get this right. Nevertheless these cookbook circuits find their way into products.

Amir may have data on many of the DACs he has tested.

 

[mdsimon2]

Here is the evidence to the contrary:

There are several ASRC chips that include an AES/SPDIF receiver but lack a digital volume control. When these ASRC chips are used in a DAC, they will precede the digital volume control and the DAC will clip intersample overs at all volume settings.

The ESS DAC chips have AES/SPDIF receivers built in, and the chip includes a digital volume control. Unfortunately the ESS chip will clip intersample overs even when the volume is turned down. Many ESS DACs will have a problem.

Given these two factors, I would expect to see plenty of DACs that have intersample clipping problems at all digital volume settings.

In many cases, you have to add additional chips to provide a digital volume control in the beginning of the signal path. The "cookbook" data sheet circuits often don't get this right. Nevertheless these cookbook circuits find their way into products.

Amir may have data on many of the DACs he has tested.

Very interesting, thank you for sharing.

Like I mentioned unfortunately I do not have any standard DACs. The MOTU Ultralite Mk5 I tested does have two ESS 9026 pro DACs but I imagine they are not using the standard AES/SPDIF receiver as it is an interface with the ability to route the digital inputs to outputs other than the DAC and wouldn't be surprised if they are using something else for volume control given the two DAC chips.

I also have an ESS 9028pro based Okto dac8 pro but I am pretty sure that also uses a separate AES/SPDIF receiver, however will still test it as I am interested to see if the volume control solves the issue.

It would be great if @amirm would add the 11.025 kHz +3 dBFS test to see if there is intersample over clipping with and without attenuation in the DAC.

Michael

 

[Marc v E]

Imo it would be also of real value if future tests of eq capable dacs and avr's are tested for problems with eq applied.

@mdsimon2 you've conducted many scenarios to test eq. Which minimal set of those would be necessary to identify problems with eq?

 

[mdsimon2]

Here are some consistent tests of the MOTU Ultralite Mk5 and Okto dac8 pro. I had previously stated the MOTU Ultralite Mk5 was clipping with no attenuation but this was not the case, rather I had incorrectly set the sensitivity on the ADC (doh!).

First the MOTU Ultralite Mk5. This is TOSLINK output at 44.1 kHz playing the same +3 dBFS 11.025 kHz that was provided by @Sokel into the MOTU and then routed to the MOTU TOSLINK output and captured by the UR23 TOSLINK to USB card. This is perfectly clean and shows the correct +3 dBFS level.

65536-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS 3.00 dBFS
-2.1 dBFS C, -0.2 dBFS A
3.0 dBFS 22 - 22k UNW
Distortion at 11,025.0 Hz, 3.0 dBFS:
THD: N/A based on 0 harmonics [20..22000 Hz]
HHD: N/A [10 .. 9]
N: -148.8 dB [20..22000 Hz]
N+D: -147.7 dBFS A
THD+N: -148.8 dB [20..22000 Hz]

Playing the same signal into the Okto but recording using the Okto as a capture device gives the same clean result.

65536-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS 3.00 dBFS
-2.1 dBFS C, -0.2 dBFS A
3.0 dBFS 22 - 22k UNW
Distortion at 11,025.0 Hz, 3.0 dBFS:
THD: N/A based on 0 harmonics [20..22000 Hz]
HHD: N/A [10 .. 9]
N: -148.8 dB [20..22000 Hz]
N+D: -147.6 dBFS A
THD+N: -148.8 dB [20..22000 Hz]

To me these results are expected as neither device has an ASRC on input.

Now looking at the DAC output, first with the MOTU Ultralite Mk5 and no attenuation in the DAC. These are recorded via a Cosmos ADC at the highest input level. The spectrum looks clean but the noise floor has raised quite a bit.

65536-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -0.57 dBFS
-5.7 dBFS C, -3.8 dBFS A
-0.6 dBFS 22 - 22k UNW
Distortion at 11,024.8 Hz, -0.6 dBFS:
THD: N/A based on 0 harmonics [20..22000 Hz]
HHD: N/A [10 .. 9]
N: -95.1 dB [20..22000 Hz]
N+D: -96.9 dBFS A
THD+N: -95.1 dB [20..22000 Hz]

Okto looks the same with a clean spectrum but elevated noise floor.

65536-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -5.79 dBFS
-10.9 dBFS C, -9.0 dBFS A
-5.8 dBFS 22 - 22k UNW
Distortion at 11,024.8 Hz, -5.8 dBFS:
THD: N/A based on 0 harmonics [20..22000 Hz]
HHD: N/A [10 .. 9]
N: -90.5 dB [20..22000 Hz]
N+D: -97.5 dBFS A
THD+N: -90.5 dB [20..22000 Hz]

And now moving on to applying -3 dB attenuation using the DAC volume control. First the Ultralite Mk5, now the noise floor drops significantly and we see the expected result.

65536-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -2.15 dBFS
-7.3 dBFS C, -5.4 dBFS A
-2.1 dBFS 22 - 22k UNW
Distortion at 11,024.8 Hz, -2.1 dBFS:
THD: N/A based on 0 harmonics [20..22000 Hz]
HHD: N/A [10 .. 9]
N: -115.1 dB [20..22000 Hz]
N+D: -118.7 dBFS A
THD+N: -115.1 dB [20..22000 Hz]

And same thing with the Okto, drop the level by -3 dB and noise floor drops significantly.

65536-point spectrum using Blackman-Harris 7 window and 8 averages
Input RMS -7.36 dBFS
-12.5 dBFS C, -10.6 dBFS A
-7.4 dBFS 22 - 22k UNW
Distortion at 11,024.8 Hz, -7.4 dBFS:
THD: N/A based on 0 harmonics [20..22000 Hz]
HHD: N/A [10 .. 9]
N: -112.8 dB [20..22000 Hz]
N+D: -121.5 dBFS A
THD+N: -112.8 dB [20..22000 Hz]

So overall these two devices behavior similarly. The receivers are able to take the +3 dBFS 11.025 kHz input without clipping, without attenuation they show an elevated noise floor but no severe clipping and applying -3 dB in the volume control lowers the noise floor to the expected level.

Michael

 

[mdsimon2]

Imo it would be also of real value if future tests of eq capable dacs and avr's are tested for problems with eq applied.

@mdsimon2 you've conducted many scenarios to test eq. Which minimal set of those would be necessary to identify problems with eq?

I think I would do something pretty simple and hopefully representative of how most people use crossovers.

1) 30 Hz -1 dBFS tone while applying a 80 Hz 4th order LPF (mimicking a sub LPF x-over)
2) 100 Hz -1 dBFS tone while applying a 80 Hz 4th order HPF (mimicking a main HPF x-over)

I would also love the tests below with no filters applied to see how the DAC handles intersample over clipping.

3) 11.025 kHz +3 dBFS tone with DAC volume control at 0 dB
4) 11.025 kHz +3 dBFS tone with DAC volume control at -3 dB

Michael

 

[Sokel]

Did a quick one,labels are self explanatory.
Usb connection DAC (analog output) into ADC.

-9db

-6db

-3db

0db

Normal 11025Hz sine

 

[mdsimon2]

Did a quick one,labels are self explanatory.
Usb connection DAC (analog output) into ADC.


View attachment 268561

-9db


View attachment 268562

-6db



View attachment 268563

-3db


View attachment 268564

0db



View attachment 268565

Normal 11025Hz sine

Nice, thanks for sharing! Is that with attenuation in your software player or the DAC?

Michael

 

[Sokel]

Nice, thanks for sharing! Is that with attenuation in your software player or the DAC?

Michael

On the dac,Khadas Tone1 normally don't have this capabilitty but I have Ian's controller so I can get into it's parameters including VC.

 

[KSTR]

IME, clipped vs. non-clipped intersample overs don't make any audible difference because the music that produces IS-overs in the first place is already "broken" at/around the points where the IS-overs do occur. I'm open to (blind) test results that show otherwise -- setting up a test which properly emulates this is quite straightforward, not relying on actual clipping in the DAC used for listening test.

The exception is digital IS-over artifacts from SRCs etc notably when they produce nasty code wrap-arounds which of course are very audible. RME's hardware SRC (some AKM chip) in the ADI-2 Pro runs with 6dB attenuation for a good reason (thanks @John_Siau for bringing this up, in the ESS context).

Obviously, IS-overs should be gracefully handled by a DAC (+ its analog circuitry) in the form of non-sticky soft clipping and some 1..3dB of headroom before clipping. Like what I see with AK4490 and AK4493 chips. Filter setting plays a big role here, too. The closer to ideal sinc(), the larger the IS-overs from the internal upsampling present in any Delta-Sigma DAC chip.

 

[KSTR]

It would be great if @amirm would add the 11.025 kHz +3 dBFS test to see if there is intersample over clipping with and without attenuation in the DAC.

As for that, I recently tested a C-Media USB DAC chip used for motherboard sound and that one clearly has a part of the digital volume control after the DAC proper and distortion did not change when reducing output volume -- and any IS-over effects won't change as well:

 

[ElNino]

IME, clipped vs. non-clipped intersample overs don't make any audible difference because the music that produces IS-overs in the first place is already "broken" at/around the points where the IS-overs do occur. I'm open to (blind) test results that show otherwise -- setting up a test which properly emulates this is quite straightforward, not relying on actual clipping in the DAC used for listening test.

This is a great point. There was a thread from 2018 where someone attempted to test for this, but it didn't get a lot of traction: https://www.audiosciencereview.com/...ation-audible-intersample-clipping-test.2231/

 

[AnalogSteph]

The ESS DAC chips have AES/SPDIF receivers built in, and the chip includes a digital volume control. Unfortunately the ESS chip will clip intersample overs even when the volume is turned down. Many ESS DACs will have a problem.

That wouldn't be the ASRC's fault though, which seems to follow the oversampling filter + volume control section in every ESS DAC block diagram I've seen. It's probably overflow (limiter) issues in the digital filter then. A number of parts common in the late '80s to early '90s seems to have been afflicted by this, including the Sony CXD1144B, CXD1244 and NPC SM5813AP - the later CXD2560 and CXD2567 were unaffected. Ironically, the parts used in antique first-gen CD players generally got it right, like the Philips SAA7030, but also the later SAA7220. (The situation among Yamaha or other NPC parts is unclear to me.)

 

[djwkyoto]

Thanks to everyone in this thread for addressing these issues, especially the OP @mdsimon2 and @John_Siau .
I happen to have a MiniDSP SHD (and, coincidentally, this particular Steely Dan CD ) and would like to know how to do a workaround for this issue of clipping intersample overs.

I have a stereo setup with a CD Player via SPDIF and a (Linux) computer via USB hooked up to the MiniDSP. Other digital sources may follow, probably via Toslink. I'm not too worried about slightly reduced SINAD (even though I use Dirac Live), so a -3dB attenuation that solves the issue would be fine with me.

Thanks for your help.

 

[RichB]

Thanks to everyone in this thread for addressing these issues, especially the OP @mdsimon2 and @John_Siau .
I happen to have a MiniDSP SHD (and, coincidentally, this particular Steely Dan CD ) and would like to know how to do a workaround for this issue of clipping intersample overs.

I have a stereo setup with a CD Player via SPDIF and a (Linux) computer via USB hooked up to the MiniDSP. Other digital sources may follow, probably via Toslink. I'm not too worried about slightly reduced SINAD (even though I use Dirac Live), so a -3dB attenuation that solves the issue would be fine with me.

Thanks for your help.

If I understand correctly, the issue is avoided by attenuating (using volume control) on the digital source.
Your CD would have to have lower the volume on the digital output.
The computer is the same, your software playback would have to have this.

If you cannot attenuate the digital stream, there will be an issue with the MiniDSP.

- Rich

 

[djwkyoto]

If I understand correctly, the issue is avoided by attenuating (using volume control) on the digital source.
Your CD would have to have lower the volume on the digital output.
The computer is the same, your software playback would have to have this.

If you cannot attenuate the digital stream, there will be an issue with the MiniDSP.

- Rich

Yes, that's my understanding as well. On a computer it's gonna be relatively easy I guess, by reducing the overall volume, so it is independent from the software you use for playing music. The trickier part is how to know what's -3db. Usually you get just your standard percentage display of the volume level, no information on level.

(I guess I could use EQ software like PulseEffects, leave the EQ flat and just reduce the output by 3dB. But that would mean another conversion to 32 bit float and back to 16/44.1 and a slight delay.)

The CD Player of course doesn't have any kind of volume control, so my question is if there's an easy way to attenuate the output between the player and the SPDIF input of the MiniDSP.

 

[holbob]

3db on a pc is volume level 82.

 

[djwkyoto]

3db on a pc is volume level 82.

I don't think it's that consistent, is it?

I tried with Alsamixer under Linux and it indeed does show the decibel attenuation.
-3dB corresponds with 88% in Alsamixer and 90% on the Sound tray icon.

Anyone got an idea about the CD player? Or am I simply stuck with the problem?

 

[Mnyb]

I resampling the output of my raspberry PI ( picore player with Sox ) to 24/96 for every source and lower the level by 3dB by using before sending it to my aging Meridian G68J .

So many raspberry pi player software usually have means to alter level or sample rate for you .

 

[mdsimon2]

I don't think it's that consistent, is it?

I tried with Alsamixer under Linux and it indeed does show the decibel attenuation.
-3dB corresponds with 88% in Alsamixer and 90% on the Sound tray icon.

Anyone got an idea about the CD player? Or am I simply stuck with the problem?

For the CD player you can use a SPDIF capture card with CamillaDSP to reduce the level and then route that attenuated output to a device with a SPDIF output. Something like a Hifime S2 Digi -> https://hifimediy.com/product/s2-digi/ should work as it has both SPDIF input and output. However in my experience I get drop outs when I use it like that. I think it is because the capture side is clocked by the SPDIF input and the output side is clocked by something else. It is possible that you may not have drop outs if you enable rate adjust / resampling in CamillaDSP to bridge the clock domain. Worst case if you have drop outs you can get a separate capture card like this -> https://hifimediy.com/product/hifime-ur23-spdif-optical-to-usb-converter/ and use the S2 Digi for digital output but in that case you will definitely need to enable rate adjust / resampling in CamillaDSP.

Michael

 

[djwkyoto]

For the CD player you can use a SPDIF capture card with CamillaDSP to reduce the level and then route that attenuated output to a device with a SPDIF output. Something like a Hifime S2 Digi -> https://hifimediy.com/product/s2-digi/ should work as it has both SPDIF input and output. However in my experience I get drop outs when I use it like that. I think it is because the capture side is clocked by the SPDIF input and the output side is clocked by something else. It is possible that you may not have drop outs if you enable rate adjust / resampling in CamillaDSP to bridge the clock domain. Worst case if you have drop outs you can get a separate capture card like this -> https://hifimediy.com/product/hifime-ur23-spdif-optical-to-usb-converter/ and use the S2 Digi for digital output but in that case you will definitely need to enable rate adjust / resampling in CamillaDSP.

Michael

Hey Michael,

thank you, that's really, really helpful, I'll look into these options.

Kind regards.