| Bernhard |
My new project:
A 16 bit DAC that uses one DAC chip per bit = 16 DACs per channel.
Chips are parallel data input / current output DACs BB PCM54.
They work as precision switchable current sources.
All 16 currents are summed up on the virtual ground of the I/V stage.
Every bit's output current is very fine adjustable, within the magnitude of 2 LSBs, having a low-ohm pot + resistor network between Iout and virtual ground.
Calibration will be done watching harmonics fade away on the analyzer.
If it works ok, there are options to calibrate without pots and resistors. |
|
|
| Rescue Toaster |
Shouldn't the error of even the MSB be less than one LSB? At least on a decent chip?
And isn't it likely you'll introduce glitch noise? |
|
|
| BrianL |
Do you think that a bunch of pots and resistors that you trim (with what instrumentation?) will be better than the precision trimming of an individual DAC? I think not.
Surely it would be easier to buy and 18 or 20 or 22 or 24 bit DAC that has linearity to better than 16 bits.
Read data sheets carefully |
|
|
| Bernhard |
| quote: | Originally posted by Rescue Toaster
Shouldn't the error of even the MSB be less than one LSB? At least on a decent chip?
|
Tolerance should be much tighter, about 1/50 LSB but it isn't.
That is why a few chips have adjustment for MSB + next 3 bits. |
|
|
| Bernhard |
| quote: | Originally posted by BrianL
Do you think that a bunch of pots and resistors that you trim (with what instrumentation?) will be better than the precision trimming of an individual DAC? I think not.
Surely it would be easier to buy and 18 or 20 or 22 or 24 bit DAC that has linearity to better than 16 bits.
Read data sheets carefully |
The precision trimming of an individual DAC is not that precise, thats why there are selection grades.
It is a simple game, all you must do is make disappear those vertical lines on the screen.
Nobody should believe that any 20bit chip is better compared to a 16 bit one.
The most expensive 18bit BB instrumentation DAC has 16bit accuracy from precise factory trim and needs user calibration with 4 pots to perhaps reach 18 bit. |
|
|
| Bernhard |
I have built one channel and it works so far.
So far means the sound is distorted, harmonics -36dB @ FS which is about 2% ??? distortion.
What I had overlooked is that audio DACs have poor gain error of 2% :dead:
Now I will put in bigger pots and see how far the DAC can be calibrated.
Later the bit errors can be compensated to one or two LSBs by choosing different codes and the fine adjustment again with the pots. |
|
|
| janneman |
| quote: | Originally posted by Bernhard
I have built one channel and it works so far.
So far means the sound is distorted, harmonics -36dB @ FS which is about 2% ??? distortion.
What I had overlooked is that audio DACs have poor gain error of 2% :dead:
Now I will put in bigger pots and see how far the DAC can be calibrated.
Later the bit errors can be compensated to one or two LSBs by choosing different codes and the fine adjustment again with the pots. |
Bernhard,
Two comments:
- the poor gain error should not cause this 2% distortion. This is a linear distortion meaning that your output level in RMS terms would be off from the design value, but should be undistorted.
- even if the gain error would cause distortion, the fact that you have 16 DACs in parallel would average it out and the result should be much better.
I think there is another mechanism causing the high distortion. How sure are you that the input signal is distortion free?
Jan Didden |
|
|
| janneman |
Too late for edit: You may check the input sigbnal and general setup by temporarily disconnecting 15 of the 16 DACs. How does that look?
Jan Didden |
|
|
| Bernhard |
Hi Jan
each DAC is only switching from BPZ to the corresponding fixed bit current.
One DAC per bit.
I use the DACs as precision current sources with integrated switch. |
|
|
| janneman |
| quote: | Originally posted by Bernhard
Hi Jan
each DAC is only switching from BPZ to the corresponding fixed bit current.
One DAC per bit.
I use the DACs as precision current sources with integrated switch. |
I still don't get the whole picture. What's BPZ?
Jan Didden |
|
|
| Bernhard |
Bipolar Zero. Zero output current.
DAC1 BPZ - MSB
DAC2 BPZ- bit2
DAC3 BPZ - bit3
.
.
.
DAC16 BPZ - LSB |
|
|
| janneman |
| quote: | Originally posted by Bernhard
Bipolar Zero. Zero output current.
DAC1 BPZ - MSB
DAC2 BPZ- bit2
DAC3 BPZ - bit3
.
.
.
DAC16 BPZ - LSB |
OK,
Apart from the fact the generally the LSB is bit 0 and the MSB is bit 15 (just convention, but it means I have to do Real Thinking!), I try to see what you did.
So, we have a digital data word you want to convert to analog. Do you connect, say, input data bit line 5 to DAC # 5, so that DAC # 5 switches its bit # 5 depending on the state of the input bit 5? Do you have a diagram?
Jan Didden |
|
|
| Bernhard |
| quote: | Originally posted by janneman
input data bit line 5 to DAC # 5, so that DAC # 5 switches its bit # 5 depending on the state of the input bit 5 |
Exactly. 55555.
The PCM54 is only for testing.
The final chips seem to have gain adjustment.
I have replaced first two pots, it is getting better. |
|
|
| janneman |
So, Bernhard, not to discourage you, but what you seem to do is take 16 DACs, use only one out of 16 current switches from each.
Then, because you lose the inherent advantage of IC matching, you need to individually adjust each current switch to try to come close to the matching you would have in a single 16 bit DAC.
Where is the advantage?
Jan Didden |
|
|
| Bernhard |
| quote: | Originally posted by janneman
you need to individually adjust each current switch to try to come close to the matching you would have in a single 16 bit DAC.
Where is the advantage?
|
That I can do it.
Integrated DAC usually have only one pot for the MSB. |
|
|
| janneman |
| quote: | Originally posted by Bernhard
That I can do it.
Integrated DAC usually have only one pot for the MSB. |
... but at any rate the integrated DAC is way better than anything you can come up with. Have you thought about a tracking temp compensation for your battery of pots, to mention just one issue?
Jan Didden |
|
|
| Bernhard |
| quote: | Originally posted by janneman
... but at any rate the integrated DAC is way better than anything you can come up with. |
Are you sure ?
In the end the adjustment range of the pots will be very tight, tighter than MSB adjustments of integrated DACs.
If necessary, I will build a master-slave configuration where the slaves compensate the gain errors of the masters to below 1 or 2 LSBs.
All lab instruments have calibration, why not DACs ? |
|
|
| janneman |
| quote: | Originally posted by Bernhard
Are you sure ?
[snip] |
No. But pretty confident that it won't work better than the standard DAC, not even close :cool:
But I will follow your adventure with interest.
| quote: | Originally posted by Bernhard
[snip]All lab instruments have calibration, why not DACs ? |
All OLD lab instruments and those that push the technological envelope have cal, because we cannot yet built them in such a way that they are accurate and stable over time.
All modern, medium-accuracy lab instruments are self-calibrating or need only one cal after production.
Nobody would take a self-calibrating or stable instrument and modify it for manual calibration.;)
Jan Didden |
|
|
| janneman |
Think about it. What you want to do is take advantage of the inherent strength of the IC process, the tight ratio matching and temp tracking in DACs. You don't want to throw that away, because you'r lost then; no matter your ambition, there's no way you can improve that with manually adjusted, variyng, temp dependent external components.
What you are looking for is a way to improve on the inherent errors in DACs. One way is to somehow combine multiple DACs so that the errors average out. Something like putting multiple transistors in parallel so that the resulting noise is less than a single transistor.
How can you do that? What comes to mind: taking multiple DACs, summing their voltage output (or current output into an I/V) will work. How much will the improvement be? Don't know.
Another way is randomizing and averaging out the errors as done in the dCS ringDACs.
Think it through, maybe there are other ways. Your way is a dead end. If someone knowlegable would look at your proposal, immediately the reaction will be: hey, if we could integrate that stuff, we would get MUCH better performance, and even won't need the adjustment stuff...
Jan Didden |
|
|
| Bernhard |
| quote: | Originally posted by janneman
If someone knowlegable would look at your proposal, immediately the reaction will be: hey, if we could integrate that stuff, we would get MUCH better performance, and even won't need the adjustment stuff...
|
That is the point. The pots can not be integrated into the DAC.
If the total adjustment range is only 1 LSB, how much can the drift be ?
The big 18 bit BB DAC729 also uses 4 optional pots to reach full resolution, which is not possible in chip production.
For temperature tracking I could mount the DACs on a common heatsink and put them in a closed chamber. |
|
|
| Bernhard |
It is much better now with 1k multiturn pots.
Even -60dB is easy to adjust and sounds very clean.
I will post some pictures later. |
|
|
| Bernhard |
| Calibration is real tricky but now all distortion artefacts are gone, -60dB sounds clean except some background pink noise. |
|
|
| Bernhard |
Next stage could be a colinear DAC.
 |
|
|
|
