Stable OpAmp With Phase Response >180°

Hi

Check page 62-65 from this document: application_note_67.pdf

The "super gain block" presented here is claimed to be unity-gain stable although it has a phase response above 180° within a certain frequency band. I run a few simultians with discrete designs modified to have similar o/l behaviour, and they showed good stabilty as well.

Now why's that? I (and many textbooks) think that Nyquist told as differently--it appears however that phase response needs to be below 180° at the unity gain frequency only. Any smart guy out here who could shed some light on this? Thanks!

Samuel

Did you plot the composite phase shift?

While I may be reading what plots are what incorrectly, their plot in fig 94 does not look like the simple sum of 92 and 93. Perhaps phase shift scale on 94 wrong (it looks like maybe 2x).

JR

Their plot in fig 94 does not look like the simple sum of 92 and 93.

Click to expand...

It's not supposed to be the sum of both, it's 2x 92 plus 1x 93. The legend for figure 94 should read "[...] (Shown in Figure 95)" I think.

Samuel

OK, in that case it looks like an oscillator above 10k. I wouldn't try to use that at unity gain. YMMV

JR

So what's the noise gain of S1/S2 in figure 96?

Samuel

Is this a test?

Noise gain of S2 is approx 2 wideband, noise gain of S1 varies with frequency but drops to unity at high frequency.

This contradicts my unerstanding of stability criteria and may be a quirk of the specific application. Stability at exactly 10k may not be an issue in a 10k oscillator (phase shift passes through 180' at 10k on their graph).

I am sill not inclined to consider this for a general purpose circuit block, while 120dB of loop gain at 10k would be attractive.

JR

Is this a test?

Click to expand...

In some sense yes--it is a test if my understanding of the Wien bridge oscillator is sound (it is at least with respect to noise gain).

This contradicts my unerstanding of stability criteria and may be a quirk of the specific application.

Click to expand...

I'm able to relate to this contradiction, that's why this thread got started.

I am still not inclined to consider this for a general purpose circuit block, while 120 dB of loop gain at 10k would be attractive.

Click to expand...

It's surely not a general purpose opamp but it would be very attractive for applications with fixed gain. Simulations indicate that with a bit of phase lag compensation things behave very well at noise gains of 20 dB and less. Above that it gets a bit peaky (peak-y) due to excessive phase shift.

Samuel

I'm not a sim guy.. I'll wait for your reports about how it sounds.

I have enough trouble with stuff I believe should work, to pursue things I don't expect to work.

JR

Fortunately enough this design has allready been implemented, so we don't need to wait:
www.tech-diy.com/TestEquipment/SuperOscillator/super_oscillator.htm
www.amplifier.cd/Technische_Berichte/Amplifier_reiner_Sinus/Amplifier_reiner_Sinus.htm
www.amplifier.cd/Technische_Berichte/Amplifier_reiner_Sinus/amplifier_clean_sine.htm

Any idea why this oscillator works--perhaps Brad could jump in here?

Samuel

I'm a bit puzzled, but this particular ruse to pull away from the rocks is not unheard of. It may require a more elaborate stability analysis than the simple gain/phase to explicate. I'm a bit preoccupied at the moment with some clients to give it much attention.

I haven't looked at it yet either, but some time ago Gerald Stanley (the Mind Behind at Crown) mentioned a patent he'd gotten on what he called remote sensing, which I finally figured out meant getting stable, impedance-reducing feedback from the load side of a class-D amp's output filter. At high frequencies too---IIRC he mentioned a damping factor of 3000 at I think 10kHz, while adding that it was more important for marketing than sonics. It sounds as if it is a comparable feat.

One remark: the distortion out of an oscillator like the LT one will eventually be limited by voltage coefficients and self-heating effects in the resistors, and dielectric losses etc. in the caps and circuitboard material. I'd guess this will kick in around 100ppb.

EDIT: As foul as my German is I still found the second reference sehr amusiert :green:

[quote author="Samuel Groner"]Fortunately enough this design has allready been implemented, so we don't need to wait:
www.tech-diy.com/TestEquipment/SuperOscillator/super_oscillator.htm
www.amplifier.cd/Technische_Berichte/Amplifier_reiner_Sinus/Amplifier_reiner_Sinus.htm
www.amplifier.cd/Technische_Berichte/Amplifier_reiner_Sinus/amplifier_clean_sine.htm

Any idea why this oscillator works--perhaps Brad could jump in here?

Samuel[/quote]

I am questioning that this might be useful as a general purpose unity gain stage.

I am still a little curious how the agc loop deals with the circuit wanting to oscillate at 10k naturally (due to gain & phase shift) without need for much if any help. I don't doubt that it works as an oscillator. I would be more surprised if it didn't oscillate with that combination of loop gain and phase shift..

JR

Perhaps the solution is that for this oscillator topology the noise gain of the amplifiers is not what one would expect at first. Couldn't it be that noise gain is unity at the oscillator frequency and much higher elsewhere? At least I do remember another Wien bridge oscillator using a not unity gain stable amplifier--check the LT1037 datasheet.

Samuel

[quote author="Samuel Groner"]Perhaps the solution is that for this oscillator topology the noise gain of the amplifiers is not what one would expect at first. Couldn't it be that noise gain is unity at the oscillator frequency and much higher elsewhere? Samuel[/quote]

That occurred to me as well as a possibility.

[quote author="bcarso"]

I haven't looked at it yet either, but some time ago Gerald Stanley (the Mind Behind at Crown) mentioned a patent he'd gotten on what he called remote sensing, which I finally figured out meant getting stable, impedance-reducing feedback from the load side of a class-D amp's output filter. At high frequencies too---IIRC he mentioned a damping factor of 3000 at I think 10kHz, while adding that it was more important for marketing than sonics. It sounds as if it is a comparable feat.

[/quote]

I have nothing but respect for Stanley et al and work I've seen come out of his team over the years. They spec damping factor on their latest class D offering at 5000:1 but only up to 100Hz so 3000 at 10k may be a little optimistic in practice. I am surprised to see a patent on this as IIRC the later generation Peavey Class D amps did something similar, to mitigate the truly horrible DF in the earlier class D amps from the huge output filter just sitting in series. Some people (like me) question the merit of DFs that high, while conceding the marketing mileage they provide.

The problems of closing that loop at HF remains and and is one valid criticism of all (?) class D amps for serious pro touring use is HF roll off at 2 ohm loading (Crown included). Users can work around this and don't often need to drive their tweeter arrays at 2 ohms, but with full range boxes it could happen. Nowadays many amps have DSP built in so it seems almost trivial to correct. If you want to invent something else, look at the input and output response and correct that in quasi real time.

I hate looking at patents. Another one of Stanley's is feeding a 16b sine wave output from an 8 bit processor (4604719)? Luckily this one is some 20 years old as I inadvertently did something similar in my little drum tuner to generate a more than 8 bit sine wave output from the 8 bit processor. It didn't strike me as an invention. Just using digital technology how you must to get stuff done in the digital domain. The concept of double precision is even built into some digital architectures, but like I said I don't mean to be critical of Gerald Stanley's contributions to the art. Whether it is all patentable or not doesn't detract from solid engineering.

JR

[quote author="Samuel Groner"]Perhaps the solution is that for this oscillator topology the noise gain of the amplifiers is not what one would expect at first. Couldn't it be that noise gain is unity at the oscillator frequency and much higher elsewhere? At least I do remember another Wien bridge oscillator using a not unity gain stable amplifier--check the LT1037 datasheet.

Samuel[/quote]

Indeed... if you look at wein bridge topology the feedback kind of nulls itself out at resonance, so noise gain is whatever. I was going to mention this earlier but it doesn't explain the second similar stage for which the negative (and should turn positive at 180') feedback is very much there in full force at 10kHz.

The only concern for this app as I see it is if you can't damp down the level of 10k oscillation with the AGC loop. I don't believe it coincidence that 180' of phase shift occurs at 10kHz.

There was at least one comment to that effect under one of your links demonstrating that it works (I don't doubt it will oscillate). Perhaps stability criteria for circuits designed to oscillate, means ability to damp response above and below oscillation. If phase shift greater than 180' makes that feedback negative again that modified definition mostly fits. The phase shift passes through 180' a second time but the loop gain is some 80dB lower at that point. It might be interesting if that second 180' point fell on a harmonic of 10k which it doesn't appear to.

It might also be interesting to chase down this circuits designer for a more complete explanation. I'm just speculating (nice way to call BS on myself).

JR

Pardon the somewhat off-topic:

[quote author="JohnRoberts"]The problems of closing that loop at HF remains and and is one valid criticism of all (?) class D amps ...JR[/quote]

Bruno Putzeys' UcD approach (wholly owned but easy to license from NXP/Philips---just use their parts, like the output FETs) inherently includes the output filter as part of the loop and works beautifully. It still has the deadtime-versus-danger tradeoff of all conventional output stages save Gerald's patented approach, and also (hotly denied as an issue by Hypex btw) the potential problem of heterodyne artifacts among nearby multiple channels, being intrinsically a variable-frequency design, but with those modest reservations it kicks ass IMO. If I can't use Stanley's (and I could only use it for stuff for Harman) I use UcD if the Fi is to be Hi.

Thanx, I'm not familar with this one but will check it out. I am a sometimes lurker on a forum Bruno moderates, another pretty smart guy to add to the list we've already mentioned.

The SOTA that I have seen in the market for multi kW class D amps still show sensitivity to speaker loading at 20 kHz. Most currently just underdamp at 8 ohms and suffer some attenuation at 2 ohms.

IMO not a big issue and easily managed but it bothers the Spec sheet readers who are accustomed to conventional analog amps not expressing this particular response deterioration with loading.

JR

US patent 6798285 seems to give more detail on this though I have to give it another read first.

Samuel

I quickly run an open-loop gain simulation of the topology and values given in the patent and got this result (assuming ideal opamps with 120 dB o/l gain and keeping the last error correction stage closed as shown): halcroamp.gif

It really looks as if the phase needs to be below 180° at unity loop gain only and not below that frequency. At least intuitively that makes sense--once you close the loop, there is--given sufficient feedback--no phase shift in the path to the inverting input. Only once feedback decreases towards zero the o/l gain/phase comes into play.

This in turn means that these sort of amplifiers is only stable below a certain noise gain.

Samuel

[quote author="Samuel Groner"]I quickly run an open-loop gain simulation of the topology and values given in the patent and got this result (assuming ideal opamps with 120 dB o/l gain and keeping the last error correction stage closed as shown): halcroamp.gif

It really looks as if the phase needs to be below 180° at unity loop gain only and not below that frequency. At least intuitively that makes sense--once you close the loop, there is--given sufficient feedback--no phase shift in the path to the inverting input. Only once feedback decreases towards zero the o/l gain/phase comes into play.

This in turn means that these sort of amplifiers is only stable below a certain noise gain.

Samuel[/quote]

Thanx for the pat reference. Very interesting.

My brain muscle is not very supple and I'm having more than a little difficulty grasping the concept of stability with that much phase shift in the audio path. As I have previously speculated, phase shift past 180' seems like it would rotate the feedback back to stable negative again so my only concern is about quirky behavior at that first pass through 180'. The slope there appears to be very steep, so frequency band will be narrow. If I were to bench test this, I would look at performance in this specific region. With modern high speed electronics if this funny region could be bumped up above 20 kHz while still delivering the huge loop gain margins for audio band, I could learn to like this approach.

This appears to be a very powerful concept, if it delivers all it claims with no real downsides beside the modest added complexity. Patent issued in 2002 so only 12 or 15 more years to go.

BTW the patent is a good read for the overview it provides on prior art wrt low distortion amp design.

JR