Rerurn to Romy the Cat's Site

In the Forum: Didital Things
In the Thread: DC offset for A/D converters.
Post Subject: DC offset for A/D converters.Posted by Romy the Cat on: 3/13/2009

Since the new revised Rohde & Schwarz EU-6201 become a leading recoding tuner and since I bound with Lavry A/D, the Lavry prosessor makes a good come back. The competition between Lavry AD122 and Pacific Microsonics never was finalized and I generally found then very competitive. Sometime I feel some advantage of Pacific and sometime I feel that it is not there. There was one subject that always bothered me: Lavry AD122 has always the lower LEDs of it meter, at -45dB and lover lit. vs Pacific is silent doe to -10000000dB. Today I realized that it is just a simple DC.

It looks like an internal DC, I wish it was not there but it is. It shall not be there for a unit of this caliber. Pacific run own slow DC removal circuit on it’s input.  Lavry does the same but at digital level, which is not good conceptually and listening assessments clearly confirm it. The good part that Lavry allow to defeat it… and it is what I was doing all time. Here is Lavry comments from pro audio forum:

FROM DAN LAVRY to ANDY KREHM on August 17/08

"The DC on is only for factory testing. The default is DC ON and I believe it is stated in the manual.

DC does not belong in audio, and in fact, I read some times back some medical reports that using audio gear at very low frequencies a few HZ can cause harm to humans. In any case, one can not hear 10Hz or 20Hz, and the main reason to go that low with DC removal is to provide a sensible margin to insure that the lows you hear are very clean.

Regarding the LavryGold AD: The converter itself does allow some DC to go through the conversion, but the DC content is filtered out AFTER the conversion. Proper operation requires the DC removal to be ON. This is a default setting and the operation manual states so. The DC OFF setting is for factory use only.

Say you have a 10V peak to peak signal. It goes all the way down to -5V and all the way up to +5V. Now say the same signal is "riding on a 1V DC". That 1V DC makes the peak go to +6V and the deep to -4V. If you try to meter that +6 to -4V signal, you will end up with a very wrong reading because +6V peak yields a very different reading the +5V peak. In fact a 1% of DC content relative to full scale will read as a -40dBFS noise floor. A 0.1% reads as -60dBFS. So reading the audio signal requires removal of the DC.

Such addition of 1VDC does not hurt the audio, as long as it does not cause the signal to clip.

Of course, one does not want to have huge amount of DC, because it would lower the converter range (clipping) on one side (signal peak) or the other side (signal deep). That would lower the dynamic range. But a few mill volts (a few percent of full scale) of DC do not matter much. The proof - LavryGold AD has 127dB dynamic range (unweighted) and around 130dB A weighted.

However, just a few millivolts of DC can ruin your metering. Audio test systems do not include DC, they have a built in DC filters, such as 10HZ or 20Hz low pass. The LavryGold AD has such a DC removal as well.

Aside from metering, it is a good idea to filter out content at very low frequencies such as 20Hz or so. There a number of reasons for that, and it would take a lot of time to explain it all.

A couple of side points that may be of interest to some:

DC is a theoretical concept. DC is the AVERAGE level over some given time. With real devices, especially semiconductors (and tubes), if you wait long enough, you realize that the level goes up and down around that average that we call DC. The longer you wait, the higher and lower the deviations from the average DC are.

Engineers classify that "behavior" as "low frequency noise". The 2 main mechanisms are:

1. "1/f noise" which simply increases as you lower the frequency, so the noise at say 5Hz is twice the noise of 10Hz...

2. "Pop corn noise", which is much less predictable, it depends on many factors in the manufacturing of semiconductors...

There are alternate names to "1/f" and "popcorn" and there are other mechanisms as well. The point is, the DC does move a lot, and in most applications, when you do not need it, it is best to filter out the very frequencies one does not need.

In audio, we are lucky - we do not hear at frequency so low, where pop corn noise and much of the 1/f noise occurs. Should one conclude that "if you do not hear it will not hurt you"? In some cases including DC will complicate matters a lot, in other cases it does not. The key is "HOW to remove DC", and "WHERE to remove DC".

Dan Lavry"

OK, I am not in position to argue with Dan in technical aspect  but I know how it sounds in context of my objective better then Mr. Lavry. So, he might insist in LF filtering or anything else but it very much not for me. So, if I would like do not have DC and do not have any ULF filtration on digital domain (!!!!) engaged then what other options do I have?

I wonder why do not inject from the source that feeds Lavry some contra-voltages to null out the AD122’s internals DC running away. I implemented in my R&S MPX processor that connected to Lavry my own analog DC compensator (Better then coupling caps or transformers) and currently the unit outputs around 0.0015VDC in each channels. How about to calibrate the DC compensator not around the absolute 0V but around the 0dB on the Lavry meter?

Romy the Cat

Rerurn to Romy the Cat's Site