John Curl's Blowtorch preamplifier part II

[janneman]

/quote
AP’s patent-pending “continuous sweep” DSP technique calculates 14 measurements in 7 seconds from a single logarithmic sweep, then graphs results of each measurement for review in the Measurement Selector.
/unquote

[scott wurcer]

Originally posted by janneman


BTW The need for carefull synchronizing etc gives you multitone components like 66.01Hz or 2,003Hz and such. Weird, but doable if your digital generator has enough bits.

jd

I've been playing around with this in spare time. Yes it's fairly trivial to generate floating point waveforms that need no windowing. Even so for large enough FFT's and the right windows the resolution is more than enough for the job.

[Bob Cordell]

Originally posted by syn08


Of course Bob, but this is still an excellent workaround to avoid some of the LT1166 HF issues. Without this cap, I have to agree with John and others that LT1166 will never achieve high performances. Of course the cap somehow defeats (at HF) the original purpose of the LT1166, but this is (according to my results) a very good trade off.

If you have a better way to avoid the LT1166 HF issues, please post a schematic and some results.

P.S. You could save some grey cells by avoiding lecturing me on the LT1166 functionality. Apparently, you haven't noticed my previous message and the link to the LT1166 internal schematic. Have you used that schematic to understand where the LT1166 issues are coming from?

P.P.S. I think the amp frequency compensation is totally irrelevant here. We are talking about minimizing the open loop HF distortions. If you mean the local LT1166 feedback compensation then I guess it's a schematic simplification. Without that compensation, the local feedback loop is totally unstable and those measurements couldn't be done.

Hi Syn08,

My intent was not to lecture you; however, my style is to be more complete in explaining things so that others less experienced reading the thread can gain some understanding. You obviously know a lot more about the LT1166 than most others here. It also sounds like you have found a workable solution to using the LT1166, but one that is different than the one I was pursuing about three years ago.

In my experience, there are two major issues to tackle with the LT1166. Bear in mind that this is in the context of my goal to use the 1166 in the way the designers presumably intended.

The LT1166 implements a common-mode bias spread control loop, where what we are calling the common mode is the bias spread. This loop forces the product of the currents of the output transistors to be constant (at a given LT1166 temperature). The stability of this loop is a big concern and is the biggest reason I had to do a lot of reverse engineering of the LT1166. In many situations, the internal compensation of the LT1166 does not keep the gain crossover frequency of this common mode loop to a low enough frequency. Parasitic oscillations thus can result. I recall that compensating this loop can be tricky. Your solution of placing a 1 uF cap across the bias spread terminals of the LT1166 solves this problem.

The other big issue with the 1166 is the way in which the conventional global feedback compensation is applied. This again largely becomes a non-issue when the 1 uF capacitor solution is used.

When the 1166 is being used as a dynamic shunt bias spreader, being fed with VAS current sources with signal (as we drive a conventional Vbe multiplier), the bias spread changes with the magnitude of the current being sourced to, or sunk from, the load. This is a necessary behavior if neither of the pair of output transistors is to be turned off, as required by the translinear operating law of the LT1166. This in turn means that the common mode spreading voltage is a highly nonlinear version of the signal. If any of this gets into the signal path in such a way that it is not canceled out, distortion will result.

This means that if conventional Miller compensation is used, and the Miller capacitor is tied to one end of the LT1166 bias spreader, considerable distortion will result. This was why I was asking about how the overall amplifier compensation was implemented. If instead, two Miller capacitors of equal value are tapped from both ends of the LT1166 bias spreader, this distortion will be greatly reduced.

If other types of non-Miller compensation are used to stabilize the global feedback loop, this may not be an issue.

Cheers,
Bob

[syn08]

Originally posted by Bob Cordell
This means that if conventional Miller compensation is used, and the Miller capacitor is tied to one end of the LT1166 bias spreader, considerable distortion will result.

This is correct. The bottom line is that LT1166 can barely be used with good results in non-symmetrical designs.

What I eventually found as an optimal is a lead-lag compensation from each the 1 and 4 terminals to the ground. Usually 47p in series with 50-100ohm will do very well.

[john curl]

Didn't I say that in the first place?

[1audio]

Originally posted by janneman
/quote
AP’s patent-pending “continuous sweep” DSP technique calculates 14 measurements in 7 seconds from a single logarithmic sweep, then graphs results of each measurement for review in the Measurement Selector.
/unquote

The chirp is a popular stimulus for the acoustic testing. It can be used with FFT and monitoring the source to get gain (response), phase and with some more math HD and IMD. Praxis has a good implementation of it. It is a very transient signal and depending on the low frequency limit and the frequency resolution is possible to get good input in 100 mS. If I wasn't swamped with other work I would try it on some amps to see if there is a difference in phase response at different power levels for different sweep times.

[scott wurcer]

Originally posted by 1audio


The chirp is a popular stimulus for the acoustic testing.

Dick Heyser pioneered the use of the chirp in conjunction with the
Hilbert transform for audio analysis. I don't think THD was ever addressed, it might make for some interesting experiments.

[andy_c]

There's an interesting article on this here (PDF file, 3 MB).

[Bob Cordell]

Originally posted by john curl
Didn't I say that in the first place?

No.

You said the LT1166 was incapable of high performance without explaining why.

Both syn08 and I have discovered and explained why, and have pointed out how to avoid these shortcomings. Syn08 described using twin lead-lag networks hanging off of either end of the LT1166 bias spreader to avoid distortion in the case where the bias spreading voltage is allowed to be a function of signal. I described using twin Miller compensation capacitors hanging off the ends of the LT1166 bias spreader.

Cheers,
Bob

[KSTR]

Originally posted by andy_c
There's an interesting article on this here (PDF file, 3 MB).

Yes, that's a good one, pretty complete.

Nice stuff here from A.Farina himself
http://pcfarina.eng.unipr.it/Public/Papers/list_pub.htm
http://pcfarina.eng.unipr.it/Public/.../134-AES00.PDF The original Paper
http://pcfarina.eng.unipr.it/Public/...cSound2007.pdf Detailing improvements to the above
http://pcfarina.eng.unipr.it/Public/...246-AES126.pdf The silence-sweep method

Some time ago I mentioned that I wanted to try swept multione (two-tone, non-identical signal levels) testing with the farina method, which I did, as a basic concept (and it works). Only that it's hard to figure out how to exactly interpret the results and what the requirements are for the best suited stimulus....

- Klaus