[sci.electronics] Digital/Analog converter help needed

b_haughey@ccvax.ucd.ie (Brian J Haughey) (07/16/90)

[Repost of a query that died somewhere...]

Hi there.

I'm planning on doing some work with D-A converters, and I need some 
info before I start.

What I am looking for is info on D-A converters that can directly drive
speakers. My query is : would the scaling factors of such a D-A have
to change as the volume went from very low to very high ?

Any advice, tips, or references welcome. Thanks a lot.

-- bjh

B_Haughey@ccvax.ucd.ie ||         "There's no future in time travel"
b_haughey@ccvax.bitnet                   

rich@island.uu.net (Rich Fanning) (07/19/90)

In article <3550.26a1d2d7@ccvax.ucd.ie> b_haughey@ccvax.ucd.ie (Brian J Haughey) writes:
>What I am looking for is info on D-A converters that can directly drive
>speakers. My query is : would the scaling factors of such a D-A have
>to change as the volume went from very low to very high ?

I'm not exactly sure what you want here.  You say "D-A converters that
can directly drive speakers".  By this do you mean a "digital amplifier"?

There are at least two ways to do this: convert D/A at a one-volt, high
impedance level, and amplify in analog.  Or amplify the digital signal pulses,
send through a filter to round off all those nasty sharp edges, and out
to the speaker.

Either way, it would probably be desirable to multiply the 16-bit audio
coming in by a scale factor which represents the "volume".
Given fast enough hardware, it would probably be reasonable to do a
"brute force" approach:  represent the volume by a 16-bit value, and multiply
the 16-bit audio signal to get a 32 bit result.  Take the top 16 bits,
and there's your scaled digital signal.

No doubt, there are more efficient ways to do it.  And a simple multiply
does not take care of rounding problems.  How do professional digital mixers
do it?
-- 
	 Rich Fanning         {sun,ucbcad,uunet}island!rich      

damos@otc.otca.oz (Daniel Amos) (07/20/90)

in article <3550.26a1d2d7@ccvax.ucd.ie>, b_haughey@ccvax.ucd.ie (Brian J Haughey) says:
> Xref: otc sci.electronics:3089 rec.audio:747
> 
> [Repost of a query that died somewhere...]
> 
> Hi there.
> 
> I'm planning on doing some work with D-A converters, and I need some 
> info before I start.
> 
> What I am looking for is info on D-A converters that can directly drive
> speakers. My query is : would the scaling factors of such a D-A have
> to change as the volume went from very low to very high ?
> 
In all my experiences with DACs, I haven't seen one that could drive
speakers directly.  Generally, DACs are designed to produce a reasonably
accurate analog representation of a digital number - they don't include
high power amplifiers to drive low impedance loads.  A possible reason
for this is that amplifiers usually dissipate a fair amount of heat,
which is not really good for resistor chains inside DACs - you would
prefer temperature stability.

In practice, (ie if heat problems could be solved), a high resolution
DAC (eg 22 bits) could be used, and a digital multiplier would multiply
the bit stream from the digital source by a constant (the volume setting),
and the DAC would output the result.  Naturally, the DAC would require a
peak-to-peak output voltage of at least 50V (to provide a reasonable
output power).

Dan Amos
||||OTC||
Services R&D

francis@cs.ua.oz.au (Francis Vaughan) (07/20/90)

In article <1839@island.uu.net>, rich@island.uu.net (Rich Fanning) writes:
|>In article <3550.26a1d2d7@ccvax.ucd.ie> b_haughey@ccvax.ucd.ie (Brian
J Haughey) writes:
|>>What I am looking for is info on D-A converters that can directly drive
|>>speakers. My query is : would the scaling factors of such a D-A have
|>>to change as the volume went from very low to very high ?
|>
|>I'm not exactly sure what you want here.  You say "D-A converters that
|>can directly drive speakers".  By this do you mean a "digital amplifier"?
|>
|>There are at least two ways to do this: convert D/A at a one-volt, high
|>impedance level, and amplify in analog.  Or amplify the digital signal
pulses,
|>send through a filter to round off all those nasty sharp edges, and out
|>to the speaker.
|>
|>Either way, it would probably be desirable to multiply the 16-bit audio
|>coming in by a scale factor which represents the "volume".
|>Given fast enough hardware, it would probably be reasonable to do a
|>"brute force" approach:  represent the volume by a 16-bit value, and multiply
|>the 16-bit audio signal to get a 32 bit result.  Take the top 16 bits,
|>and there's your scaled digital signal.
|>
|>No doubt, there are more efficient ways to do it.  And a simple multiply
|>does not take care of rounding problems.  How do professional digital mixers
|>do it?

Uh No. If you take the top 16 bits off a 32 bit result you will only get
the full 16 bit resolution of the source material at full volume. If you
play the system at ordinary listening levels (say 30db down) you will end
up discarding nearly half the bits. Leaving you with say 9 bits of music
and 5 bits of 0s. This would result in a drastic reduction in quality.
The Denon Audio Test disk demonstrates this very well. They have a sequence
of tracks of orchestral music recorded on the disk at differing levels of
attenuation. If you normalise the volume on playback for the different tracks
the difference in sound quality is remarkable. Tracks that are recorded 60db
down sound worse than telephone lines. The difference between 0db and 10db
is greater than the difference between a high end CD and a mid to low range
cassette player. 

To do what you suggest you would need a 32 bit DA. Some DA converters allow
you to supply an external reference voltage which coresponds to FFFF. If 
you vary this the output is essentially scaled, but the range of useful
voltages
is typicly only 3:1 and therefore of little use as a volume control. 

Other factors work against you in building a power DA. You need a
reconstruction
filter on the output. On good CD players they will often use Bessel or similar
filters. These loose power. You would have to make a filter capable of taking
the full output power of the DA and you would not want to compromise other
factors of the output stage (like output impeadance). Bessel would be rightout.
It would also be very expensive, a similar exercise to building passive speaker
crossovers.

Building a Class-D output stage may be possible but these have always sounded
really bad to average. They are pulse width modulated switches. Getting rid
of the hash and intermodulation products is still a big problem.
Hence the reconstruction filter. If you want to build a digital speaker you are
really still left with an analogue ouput stage and analogue attenuation after
the DA. 

							Francis Vaughan

bill@vrdxhq.verdix.com (William Spencer) (07/20/90)

in article <3550.26a1d2d7@ccvax.ucd.ie>, b_haughey@ccvax.ucd.ie (Brian J Haughey) says:

> What I am looking for is info on D-A converters that can directly drive
> speakers. My query is : would the scaling factors of such a D-A have
> to change as the volume went from very low to very high ?

AND most important:
> Any advice, tips, or references welcome. Thanks a lot.

What are you trying to accomplish? Low cost? Amplifier chips start about 
50 cents. This does bring up a good subject though... first, no, it's hard
enough to do 16 bit D/A, much less put power handling into the same chip.

But a higher output D/A would be nice. Best would be a high voltage output.
This output would be essentially passive: a resistor ladder connected to the
output by (FET) switches. Digital volume contol could be used but
controlling volume in analog would be better. Passive filters could be used.
If the voltage was high enough no voltage amplification would be required.
Current amplification would be needed in ther form of unity gain buffers.
Merely increasing the voltage ratings of a D/A chip is very possible.

If digital could provide a means of eliminating almost all analog 
circuitry this would answer many audiophile concerns. Digital is
complicated enough on its own without following it with complex analog.

A "digital speaker" has been built (Meridian). All it really eliminates
is analog cabling. This is not enough to justify it, especially if
there's any chance of using analog sources.

ANOTHER POSSIBILITY: "One bit" conversion at power. This is simialar 
to a "switching"  or "class D" amplifier but there's also big differences.
Despite the high effieciency and power output possible this type of 
amplifier is not popular and for good reason. Analog type switching
amplifiers rely on loop feedback. "One bit" D/A conversion if a forward
only process. Don't know what this would do with a complex load though.
"One bit" conversion could also be used for the high voltage D/A 
concept above.

Hope someone can benefit from these ideas. I'm not finished mucking
with conventional analog yet.

bill S.

rich@island.uu.net (Rich Fanning) (07/21/90)

In article <1168@sirius.ucs.adelaide.edu.au> francis@cs.ua.oz.au writes:
>Uh No. If you take the top 16 bits off a 32 bit result you will only get
>the full 16 bit resolution of the source material at full volume.

Exactly the point, at full volume.  You get out exactly what you put in.

>If you
>play the system at ordinary listening levels (say 30db down) you will end
>up discarding nearly half the bits. Leaving you with say 9 bits of music
>and 5 bits of 0s.

True, but this is exactly what happens when the engineer does a fade on the
mixing board.  The signal decreases in amplitude, and the digital values
grow smaller, leaving lots of zeroes at the top.

>The Denon Audio Test disk demonstrates this very well. They have a sequence
>of tracks of orchestral music recorded on the disk at differing levels of
>attenuation. If you normalise the volume on playback for the different tracks
>the difference in sound quality is remarkable. Tracks that are recorded 60db
>down sound worse than telephone lines.

I can only reason, not having the disk, that the recording was not made using
dither.  This will cause obvious distortions in the kind of tests you describe.
A properly dithered recording will simply have random noise added. Of course,
it won't sound better as the attenuation increases, but it ought to sound
much like the original with just noise added.

>To do what you suggest you would need a 32 bit DA. Some DA converters allow
>you to supply an external reference voltage which coresponds to FFFF. If 
>you vary this the output is essentially scaled, but the range of useful
>voltages
>is typicly only 3:1 and therefore of little use as a volume control. 

Professional digital mixers, from what I hear, can sound excellent.
And they only store 16 bits after the mix is complete.  No doubt internally
they use 32 bit accumulators to help prevent rounding errors, but the signal
as it's recorded is still only 16 bits.  

From this, I reason that a turning down a digital volume control which
emits 16-bit codes to the D/A is the same as a fade on a CD.  If you cannot
hear artifacts when a song fades out on a CD, then a digital volume control
driving a 16-bit D/A ought to be adequate.
-- 
	 Rich Fanning         {sun,ucbcad,uunet}island!rich      

francis@cs.ua.oz.au (Francis Vaughan) (07/23/90)

In article <1847@island.uu.net>, rich@island.uu.net (Rich Fanning) writes:
|>In article <1168@sirius.ucs.adelaide.edu.au> francis@cs.ua.oz.au writes:
|>>Uh No. If you take the top 16 bits off a 32 bit result you will only get
|>>the full 16 bit resolution of the source material at full volume.
|>
|>>The Denon Audio Test disk demonstrates this very well. They have a sequence
|>>of tracks of orchestral music recorded on the disk at differing levels of
|>>attenuation. If you normalise the volume on playback for the
different tracks
|>>the difference in sound quality is remarkable. Tracks that are recorded 60db
|>>down sound worse than telephone lines.
|>
|>I can only reason, not having the disk, that the recording was not made using
|>dither.  This will cause obvious distortions in the kind of tests you
describe.
|>A properly dithered recording will simply have random noise added. Of course,
|>it won't sound better as the attenuation increases, but it ought to sound
|>much like the original with just noise added.

No. Dither only helps remove some scratchy sounding stuff that corresponds to
one bit of rounding. Dither cannot help with more than a one bit rounding.
The noise added to cover the lost bits as you describe would require that if
45 db of attenuation was needed (not a lot by the way) then the signal would
occupy 8 of the 16 bits and the dither noise would also need 8 bits.
A signal to noise of exactly 0db. Not exactly High Fidelity.

The whole point is that at about -45db the signal is only 8 bits wide
(if it was
all the way up to FFFF on peaks, even fewer bits otherwise). This is getting
down to telephone dynamic range (you still have the frequency response of CD)
and the distortion is comensurate.

|>Professional digital mixers, from what I hear, can sound excellent.
|>And they only store 16 bits after the mix is complete.  No doubt internally
|>they use 32 bit accumulators to help prevent rounding errors, but the signal
|>as it's recorded is still only 16 bits.  

True, but they are fading with respect to your current listening levels. 
That is FFFF is still the loudest you want to listen to, not the loudest that
your system can ever reproduce, and they attenuate wrt to that.
The digital fader board has attenuated the signal but because
you are listening at an analogue attenuated level the distortion products are
also attenuated by the same amount, and not as noticable. 

|>From this, I reason that a turning down a digital volume control which
|>emits 16-bit codes to the D/A is the same as a fade on a CD.  If you cannot
|>hear artifacts when a song fades out on a CD, then a digital volume control
|>driving a 16-bit D/A ought to be adequate.

Again wrong. 

Sanity check time. What you imply is that an arbitrarily truncated 16 bit wide 
signal sounds as good as the origonal signal. This is patently false.

You are assuming you cannot hear the distortion artifacts on a fade out. That
was the whole point of the example with the test disk. If they are normalised
you can. This is what would happen in your proposed system.


							Francis Vaughan.

ncpjmw@brahms.amd.com (Mike Wincn) (07/25/90)

In article <1839@island.uu.net> rich@island.uu.net (Rich Fanning) writes:
>In article <3550.26a1d2d7@ccvax.ucd.ie> b_haughey@ccvax.ucd.ie (Brian J Haughey) writes:
>>What I am looking for is info on D-A converters that can directly drive
>>speakers. My query is : would the scaling factors of such a D-A have
[...]
>There are at least two ways to do this: convert D/A at a one-volt, high
>impedance level, and amplify in analog. 

Well, yeah, maybe... though I haven't heard of any 100w DACs anywhere yet.

>Or amplify the digital signal pulses,
>send through a filter to round off all those nasty sharp edges, and out
>to the speaker.

THIS idea hasn't a snow-balls chance in hell of working!  Think! ...how in
hell do propose to sum up all those digital pulses to reconstruct the 
analog program material before sending it to the load?

>Either way, it would probably be desirable to multiply the 16-bit audio
>coming in by a scale factor which represents the "volume".
>Given fast enough hardware, it would probably be reasonable to do a
>"brute force" approach:  represent the volume by a 16-bit value, and multiply
>the 16-bit audio signal to get a 32 bit result.  Take the top 16 bits,
>and there's your scaled digital signal.

Indeed, though one might have an easier time of it if one started with the
16 bits as the full range of a DAC (max output), and shifted towards the LSB 
end for attenuation, though that will only give you a factor of 2 control.  For
more precise control, amplitude interpolation will be necessary.

Mike Wincn
(408) 749-3156
ncpjmw@brahms.AMD.COM

crs@lambda.UUCP (Charlie Sorsby) (08/02/90)

In article <1990Jul24.215643.22718@amd.com>, ncpjmw@brahms.amd.com (Mike Wincn) writes:
> In article <1839@island.uu.net> rich@island.uu.net (Rich Fanning) writes:
>>In article <3550.26a1d2d7@ccvax.ucd.ie> b_haughey@ccvax.ucd.ie (Brian J Haughey) writes:
>>>What I am looking for is info on D-A converters that can directly drive
>>>speakers. My query is : would the scaling factors of such a D-A have
> [...]
>>There are at least two ways to do this: convert D/A at a one-volt, high
>>impedance level, and amplify in analog. 

> Well, yeah, maybe... though I haven't heard of any 100w DACs anywhere yet.

Whoa!

What 100w DACs?  "one volt, high impedance" doesn't imply 100w at
all.  My reading of the above suggests that the power comes from
the analog amplification.  Granted that doesn't address the
original question of driving speakers directly from DACs, but....

That isn't to say that DACs *couldn't* be built to drive
speakers...

To drive speakers directly, you would need DACs able to provide
from a few hundreds of milliwatts to a few hundreds of watts, in
which case, the criticism may be valid when speaking of commercial,
existing, DACs.

>>Or amplify the digital signal pulses,
>>send through a filter to round off all those nasty sharp edges, and out
>>to the speaker.

Filtering won't do it.  The output of a DAC *isn't* pulses; it's a
level corresponding to the *value* of the digital data.

> THIS idea hasn't a snow-balls chance in hell of working!  Think! ...how in
> hell do propose to sum up all those digital pulses to reconstruct the 
> analog program material before sending it to the load?

In principle, the same way that one does so at micro-milli-watt
levels in any other DAC application.

>>Either way, it would probably be desirable to multiply the 16-bit audio
>>coming in by a scale factor which represents the "volume".
>>Given fast enough hardware, it would probably be reasonable to do a
>>"brute force" approach:  represent the volume by a 16-bit value, and multiply
>>the 16-bit audio signal to get a 32 bit result.  Take the top 16 bits,
>>and there's your scaled digital signal.

Use a "multiplying DAC" where the analog voltage input represents a
funcition of the volume and the digital input represents the signal
and the analog output represents the scaled analog.

The problem remains that (to my knowledge) no commercial DAC is
available that will drive other than *very* small speakers, if any,
directly.

NOTE: that my disagreement is not that 100 w DACs are not
commercially available but that the point where the criticism is
applied is non sequitur in that, at that point in the existing
suggestions, a 100 w DAC is not needed.

Charlie