ESS Sabre Reference DAC (8-channel)

[Russ White]

Originally posted by schro20
I always thought that digital volume control was "bad" and all volume control should be done in analog. Reason: you only have so many bits, so if you digitally scale values from (let's say) 0-255 to go only from 0-15 you have only 16 different levels left when you started with 256. If you scale in analog you still have 256 levels between 0 and 15. The latter presumably having more fidelity. Is that argument missing the point?

(I hope not to unleash a big argument here... Just point me someplace I can educate myself.)

Or is this something that's special about the ESS chip that it can do this the "right" way and the above argument is still correct in a more generic context?

peter

Hi Peter,

Here is what I think, though I am sure some will disagree.

If you have a DAC with very good DNR its very difficult if not impossible to do better than use its built in attenuation.

If you attenuate on the digital side then you put the onus on the DAC to have great DNR. The Sabre has this in spades, so its a perfect candidate.

Show me an analog attenuation circuit that can match to 0.005dB or so across channels and still make a -132 DNR (that's about -250nV integrated across 20kHz when compared to 2Vrms, or roughly 1.78nV/root Hz). It's just not likely to happen.

That said I use analog volume control too, but when I do I use something like my Joshua Tree stepped attenuator which has very good channel matching, but not anything close to .005db .

There is nothing "wrong" with either approach. Its mostly going to come down to user preference. It's just different degrees of goodness.

My personal experience is that using the ES9008 volume control yields exceptionally good results. Better than any analog attenuation I have heard yet. Keep in mind, to get the best possible DNR (by reducing the noise floor) a balanced output is ideal when using the DAC volume control as there is nothing down stream to attenuate noise. The good news, is there is none to be heard at all in my setup. The color of the music seems to emerge from a perfectly clean and clear canvas. That's what I love about this DAC.

Cheers!
Russ

[Javin5]

If I remember correctly, when in stereo mode, the sabre chip internally parallels 4 DACs for each channel. I have searched, but found no info on the following, probably trivial questions.

(a) Does this mean that the signals at all 4 outputs are equivalent and it doesn't matter which one I use?

(b) How is the internal paralleling done?

(c) Could summing the 4 outputs somehow be done outside the chip? If so, could this potentially improve sound quality? Maybe by summing the 4 currents in the I/V converter, or, in the most extreme case by using a total of 4 I/V-converters per channel?

[eclectic2k]

Originally posted by schro20
I always thought that digital volume control was "bad" and all volume control should be done in analog. Reason: you only have so many bits...

Digital volume control *can* be bad if badly implemented. Say without proper bit-depth for the processing or no/bad dither techniques. It got quite a bad rap from some poor implementations in the PC software world among others.

here's a good start:
http://extra.benchmarkmedia.com/wiki...volume_control
(you might also find good info in app notes and such from MFG's or finding some material on DSP techniques. I've been meaning to study up on this myself, but time is limited ;-)

It would seem from all accounts here that the Sabre does a fine job of it.

-Chris

[Beefy]

Originally posted by Russ White



Oh yes the voltage itself is just fine, but the LCBPS is not designed for center tapped transformers. That said, it could easily be modified to support them.

Cheers!
Russ

Check! Centre tapped transformers always do my head in

[rossl]

Originally posted by eclectic2k


Digital volume control *can* be bad if badly implemented. Say without proper bit-depth for the processing or no/bad dither techniques. It got quite a bad rap from some poor implementations in the PC software world among others.

here's a good start:
http://extra.benchmarkmedia.com/wiki...volume_control

The volume control in the Sabre DAC has been designed so that it won't have the distortions that are discussed in the link.

I haven't designed volume control into my Sabre DAC because I'm sending full volume analog signal to a preamp, so the rest of this post is just conjecture.

A volume control, simply stated, will just shift bits down in magnitude. If you have a 16 bit signal, you can turn it down and the 16 bits will shift down until the top 8 bits are all zero, the 16 bits of data now reside at bits 9 to 24 and you will not lose any resolution if you have a DAC with a noise floor at -144dB.

Then if you turn it down some more you will lose bits of resolution. If you have a 24 bit signal, you will lose bits of resolution with any turn down of the volume setting.

So, if you have a setup with your amp where you have to turn down the volume way down all the time, it will not be good. You will always lose bits of resolution. A lot of amps are this way with way too much gain.

If you can adjust the gain of the amp so that the loudest you ever listen will correspond to the DAC volume control full up, then that would be ideal. Now a typical listening session would have the DAC volume control only reduce the signal by 6dB to 12dB.

If I were to implement a volume control, I would try it by not allowing a reduction of more than 18dB. That would correspond to a shift down of 3 bits on the signal. That would mean that 24bit data would lose 3 bits of signal, and 16 bit data would be shifted down to bits 4 thru 19. Therefore, If you have a -120db noise floor, you will still have 17 bits of signal above the noise.

This would be implemented by setting Sabre registers 0 to 7 and restricting them by only allowing a range from 0x00 down to 0x18 or 0x24.

There's a good experiment for someone to try!

[rossl]

Originally posted by Javin5

(c) Could summing the 4 outputs somehow be done outside the chip? If so, could this potentially improve sound quality? Maybe by summing the 4 currents in the I/V converter, or, in the most extreme case by using a total of 4 I/V-converters per channel?

Answer C is what we are doing. Look at the ESS web site for the schematic for the Sabre8 stereo demo board and a white paper describing the internal architecture of the part.

http://www.esstech.com/techsupp/drivers.shtm

[soundcheck]

Hi.

I am running a digital volume control (64bit float) on my PC even on 16bit material . I am allowing not more than 12db attentuation.
After all, this is sounding ways better then any preamp (passive / active) I tried before.
DigitalVolumeControl with direct DAC-Amp coupling get's you lot of problems solved.
No impedance mismatches and variations, coupling caps, cables, asf.

The key parameters to consider when talking about digital volume control are IMO more then mentioned before. You need to dimension you entire chain to be able to have lowest impact on the digital side. You need to look at the voltage of your DAC output stage, the amp gain, the speaker sensivity and your typical listening volume to get digital volume control going at lowest impact.
Keep also in mind that you easily can have 12db difference on the
music material depending on the recording level, which is increasing your "working" range to the disadvantage of digital volume control.

The best way would be IMO some kind of passive shunt-to-ground volume control on the DAC output (see also how the Lessloss guys are doing it). Unfortunately my current DAC is not really supporting this.

I'll soon face the same challenge with the Buffalo dac.

I am wondering if a passive I/V conversion on the DAC out
would be recommended? Did anybody try this?

[Goto]

I have just received my Buffalo and rigged up a passive output using 10uf Evox Rifa film caps. It is not plugged into my main system yet (tomorrow) but through an old mixing desk into headphones sounds full of potential.

If I can figure out how to modify my RBroer solid state I/V stage I shall try that next. (OK, already tried marrying the two togther, with poor results due to differences in the output voltage/current of the Buffalo and Monica chips)

All good learning, I am sure.

Mark

[hirez69]

Hi Dusfor99 (or whoever can help me),

i need you help for a problem with ESS Sabre GUI.

I'm owner of an ESS Sabre EVB. I have a problem with some
registers setting through the most recent release 2.0.2.1. (but the problem there is also with the previous version) of Sabre8 GUI (2.0.2.1.): I was not able to find in any section of the GUI the window called "Input Settings (Quantizer Setting)" where to editing the number of bits each quantizer will use. While setting in stereo mode the DAC (2 Channel) i cannot set the value of the quantizer. Through the "List all register" button I see the Register 15 setting to 00000000 (6 bits quantizer), but i cannot find the "Quantizer Settings (Input)" Window where to modify the value to 7, 8 or 9 Bits. I attach the screen capture taken from the Sabre8 Gui Help file show the windows that I cannot find.
Oherwise could you sugest another way to modify those values?
Also I don't find where to midify the "differential" from "True" to "Pseudo". I use the EVB with I2S input.

Sabre firmware is updated to the latest version available in
http://www.esstech.com/techsupp/drivers.shtm#sabre.

Thank you.

[Beefy]

Originally posted by Spartacus
[B]With music at full volume via either a media player app or the windows mixer, I can hear a low level distortion that sounds like clipping.
When I turn the volume down on the PC (and up on my pre-amp to compensate) it gets worse.

I really don't know much about digital processing..... but I do know that Windows can do some weird things when changing volume in the digital domain. I would suggest looking into something like ASIO to get a bit perfect output, or try testing from a non-PC source.

I don't know why it would show up on the Buffalo, but not Opus though