Why Don't High SINAD Receivers Exist?

[mdsimon2]

I surely am, that's why I wrote 'simply' in quotes
Not a DSP expert but I can see car-processors doing 8-16 channels DSP at 192kHz for about $4000.
The $4k AVR/Ps can barely do 48kHz DSP, use much cheaper DACs and measure much worse. No need for a DSP doctorate to see the problem here.

Pretty easy to do 8+ channels at 192 kHz if you are using IIR, things change when you start talking about FIR and room correction. Perfect example is the miniDSP C-DSP 8x12, without Dirac it has 12 channels at 192 kHz, if you upgrade to the Dirac version (which has the same exact internal hardware) that changes to 48 kHz.

As was suggested earlier you can definitely increase the number of DSPs to increase sampling rate but the main point I am making is that it is not just a matter of implementing something new on existing hardware (even for these car audio DSPs), you actually need an increase in processing power. Personally increasing the DSP sample rate is not high on my wish list but YMMV.

Michael

 

[rhollan]

I’ve a Denon AVR X2400H which connects to my TV with HDMI Arc. Will I benefit from using the spdif optical connection to my tv instead? Will i benefit from adding some of the under 100 USD Dac tested here?

Well, stereo may benefit from a better DAC, but you're likely getting it from something other than the TV. And 5.1 over SPDIF will still be as compressed as over ARC. So, unless you are taking strictly stereo audio from the TV, my gut says no.

 

[rhollan]

Regarding "taps", long FIR filters can't be used in an AVR/P because it would introduce too much lag. Low frequency resolution can be increased even with a lower tap count using some tricks. Look at what Audyssey did with XT32.



An AVR/P does more than a "car-processor" and more development effort is required. There's also additional licensing fees that need to be covered (Dolby, DTS, Auro, HDMI). There's a reason why Trinnov charges a bit more than 4k

Yes, but if you remove post-decoding processing (like Dirac Live) to another device, you can make it, and D/A conversion as good as you like (and Trinnov does it badly as far as D/A goes). I'd argue that's where the bulk of the DSP power is needed. All one should need is an industry standard protected LPCM digital audio path, and HDMI with HDCP all ready offers this.

 

[Kustomize]

Well, stereo may benefit from a better DAC, but you're likely getting it from something other than the TV. And 5.1 over SPDIF will still be as compressed as over ARC. So, unless you are taking strictly stereo audio from the TV, my gut says no.

I was referring to amirs topic where he said spdif is better than hdmi. Is that not for me? For stereo listening? I have a 5.1 consisting of Zaph DIY Audio speakers. For 2 channel listening, is HDMI just as good?

 

[rhollan]

I was referring to amirs topic where he said spdif is better than hdmi. Is that not for me? For stereo listening? I am a 5.1 consisting of Zaph DIY Audio speakers. For 2 channel listening, is HDMI just as good?

It is. For stereo. Other audio will be compressed rather defeating the benefits.

 

[markus]

Yes, but if you remove post-decoding processing (like Dirac Live) to another device, you can make it, and D/A conversion as good as you like (and Trinnov does it badly as far as D/A goes). I'd argue that's where the bulk of the DSP power is needed. All one should need is an industry standard protected LPCM digital audio path, and HDMI with HDCP all ready offers this.

Not sure I understand what type of device(s) you have in mind. What is "it"?

There is a limit how good you can make D/A conversion. The problem is the amount of bits you have to throw away when filters and bass management are applied. Not a problem as long as you're within the digital domain. The problem occurs right at the DAC where you have to take the level relationship between all channels into account. Here a loss of 20dB and (a lot) more is standard.

Having said that, there are indeed methods that could optimize available dynamic range. They haven't found their way into devices yet though.

 

[rhollan]

Not sure I understand what type of device(s) you have in mind. What is "it"?

There is a limit how good you can make D/A conversion. The problem is the amount of bits you have to throw away when filters and bass management are applied. Not a problem as long as you're within the digital domain. The problem occurs right at the DAC where you have to take the level relationship between all channels into account. Here a loss of 20dB and (a lot) more is standard.

Having said that, there are indeed methods that could optimize available dynamic range. They haven't found their way into devices yet though.

I want to separate D/A from multichannel immersive audio decoding. So, I can selectively license the multichannel immersive formats I want in one device and have excellent D/A in another, with a protected LPCM path between them, with a possible speaker and/or room correction phase inbetwern as well.

Yes, room correction will (a) distort (arguably less than the room) and (b) eat dynamic range. But a 115 dB SINAD DAC would be better than an 85 dB one in that case.

 

[markus]

I want to separate D/A from multichannel immersive audio decoding.

In order to maximize SNR you actually want the opposite because you need some kind of mechanism that can inform the D/A conversion and following gain about the processing that happened upstream. For example, how would you handle LFE which is potentially 10dB louder than all other channels? If you don't know anything about the signal/processing upstream you would need to throw away 10dB SNR on other channels just to make sure your subwoofer output doesn't clip.

So, I can selectively license the multichannel immersive formats I want in one device and have excellent D/A in another, with a protected LPCM path between them, with a possible speaker and/or room correction phase inbetwern as well.

Where would bass management happen in that scenario?

 

[rhollan]

I don't follow. Bass management can be either integrated with D/A or upstream of it. Or, are you arguing for preserving computed values greater than 0dBFS (say using 32 bit math for 24 bit values) somehow by adjusting the sub VCA? To my knowledge no integrated processor does that: they just eat SNR and reduce the level of all channels so the sub does not digitally clip over 0dBFS.

 

[markus]

I don't follow. Bass management can be either integrated with D/A or upstream of it. Or, are you arguing for preserving computed values greater than 0dBFS (say using 32 bit math for 24 bit values) somehow by adjusting the sub VCA? To my knowledge no integrated processor does that: they just eat SNR and reduce the level of all channels so the sub does not digitally clip over 0dBFS.

Guess you just explained why high SINAD isn't achievable in a bass managed system with room correction and speaker management unless there's "something" happening after D/A

By the way, bass management can never be downstream of room correction and speaker processing (PEQ, trim, delay) or it will create distortion.

 

[rhollan]

Guess you just explained why high SINAD isn't achievable in a bass managed system with room correction and speaker management unless there's "something" happening after D/A

By the way, bass management can never be downstream of room correction and speaker processing (PEQ, trim, delay) or it will create distortion.

Ah, I am willing to yield SNR (and thus SINAD) for processed audio. And yes, bass management should happen before room correction: it is not good to sum corrected signals. But I want the best SINAD starting point. And, I want bitstream decoding to LPCM based om room geometry to occur separately. Or, are you saying that room geometry (necessary for immersive format decoding) must be handled as part of room correction, and that therefore immersive format decoding must involve a (mixing) step AFTER room correction? That strikes me as contradictory.

 

[rhollan]

Guess you just explained why high SINAD isn't achievable in a bass managed system with room correction and speaker management unless there's "something" happening after D/A

By the way, bass management can never be downstream of room correction and speaker processing (PEQ, trim, delay) or it will create distortion.

But that "something" is a constant factor as LFE is recorded 10 dB lower and not a function of bass management (though every part of the process of mixing channels needs to know about their relative amplitudes, yes).

 

[markus]

Ah, I am willing to yield SNR (and thus SINAD) for processed audio. And yes, bass management should happen before room correction: it is not good to sum corrected signals. But I want the best SINAD starting point. And, I want bitstream decoding to LPCM based om room geometry to occur separately. Or, are you saying that room geometry (necessary for immersive format deciding) must be handled as part of room correction, and that therefire immersive format decoding must involve a (mixing) step AFTER room correction? That strikes me as contradictory.

The Dolby/DTS/Auro decoder needs to know how many (and which) speakers there are. Otherwise it wouldn't know what to do.

 

[markus]

But that "something" is a constant factor as LFE is recorded 10 dB lower and not a function of bass management (though every part of the process of mixing channels needs to know about their relative amplitudes, yes).

That's exactly the problem. It's NOT a constant factor when bass management is involved. The peak level in the redirected subwoofer channel (which also gets the LFE) changes with the number of channels of the source material, the number of speakers in the system and the program material.

 

[rhollan]

The Dolby/DTS/Auro decoder needs to know how many (and which) speakers there are. Otherwise it wouldn't know what to do.

But it does. I never said it didn't (or couldn't).

 

[rhollan]

That's exactly the problem. It's NOT a constant factor when bass management is involved. The peak level in the redirected subwoofer channel (which also gets the LFE) changes with the number of channels of the source material, the number of speakers in the system and the program material.

It is constant in that it does not vary with time as part of the signal. And HDMI certainly indicates channel number and there are well-known assignments of channel number to speaker.

 

[markus]

It is constant in that it does not vary with time as part of the signal. And HDMI certainly indicates channel number and there are well-known assignments of channel number to speaker.

You mean it doesn't vary when you assume the worst case scenario, i.e. every channel contains a 0dBFS signal? Have you done the math how big of a headroom you would need to reserve for that case? That's how much dynamic range you lose without even having considered headroom for speaker management and room correction.

P.S. Channel assignments can change with input format.

 

[rhollan]

You mean it doesn't vary when you assume the worst case scenario, i.e. every channel contains a 0dBFS signal? Have you done the math how big of a headroom you would need to reserve for that case? That's how much dynamic range you lose without even having considered headroom for speaker management and room correction.

P.S. Channel assignments can change with input format.

Worst case?

For 7.1 channels, about 10 dB.

And, channel format is communicated over HDMI, so I fail to see the issue. It is a given that bass management or other processing with gain will eat dynamic range. You have 144dB of it to start with for 24 bit audio, 115 dB before you sink into the noise on the best of DACs. Better than 85 or even 100 dB.

 

[markus]

Worst case?

For 7.1 channels, about 10 dB.

And, channel format is communicated over HDMI, so I fail to see the issue. It is a given that bass management or other processing with gain will eat dynamic range. You have 144dB of it to start with for 24 bit audio, 115 dB before you sink into the noise on the best of DACs. Better than 85 or even 100 dB.

You're forgetting about the redirected bass from other channels. In the case of 7.1 you end up with 20.2dB required headroom. Then add 10dB for speaker trims and 10dB for EQ and your headroom requirement becomes >40dB.

Now if you send that data via the limited capabilities of HDMI you're making your system worse and not better. Better would be to keep the 32bit audio data right up until the DAC.

 

[lashto]

Pretty easy to do 8+ channels at 192 kHz if you are using IIR, things change when you start talking about FIR and room correction. Perfect example is the miniDSP C-DSP 8x12, without Dirac it has 12 channels at 192 kHz, if you upgrade to the Dirac version (which has the same exact internal hardware) that changes to 48 kHz.

As was suggested earlier you can definitely increase the number of DSPs to increase sampling rate but the main point I am making is that it is not just a matter of implementing something new on existing hardware (even for these car audio DSPs), you actually need an increase in processing power. Personally increasing the DSP sample rate is not high on my wish list but YMMV.

Michael

The car is a room and those car-processors do "room" correction, there is even a Dirac version for cars. Can't see any difference there.
And yes a bigger DSP chip might be needed for 192kHz but so what? If you ask $5-10K or more for a AVR/P, at least put the best chips in it. It's not like those DSP chips cost a fortune, e.g. with 2x of these boards you have a 16 channel AVP for a BOM below $1000 (including case, power, connectors, etc).

In would agree that there is no need to process audio samples over 20kHz freq. But I just don't want the downsample "feature". The current situation is quite ridiculous: one has to pay a premium for 192kHz tracks, just to have them downsampled to 48kHz by a so called "highres" AVR/P. That's a wonderful (double) scam.