I don't think you will gain anything with 1) 2) 3) or 6)
...
But you can't do 7) Omitting your D6 gives really nasty overload. BAV199 won't do much to help. It's not leakage but modulation of the junction capacitance that introduces the distortion. Unfortunately, ANY diode in that position will give THD only about that of the original 990.
According to LTSpice, both 2) and 3) improve matters - not much, maybe a few dB each, but mostly orthogonal, so they add up. The caveat: my THD20 sims are at unity gain, so the effect of the Vds modulation on a JFET LTP CCS is more pronounced (Vds swings are of the same amplitude as the output), so the LTP CCS cascode definitely helps. I'm not so sure about the mechanism of the distortion reduction with Wilson CM, but it probably reduces H2 a bit. You're welcome to check both in LTSpice.
Regarding 7), almost all the Spice diode models on the net seem to be broken with respect to reverse leakage and reverse-biased junction capacitance. However, I did find a Zetex BAV99 (not BAV199) model which does give order-of-magnitude lower reverse-leakage, both DC and incremental, than a 1N4148 or similar, but YMMV - it may be broken too. I got significantly lower THD20 with this model.
Another possibility that occurred to me is to use two high reverse-breakdown voltage diodes in series - e.g. MMBD7000 or similar, for the anti-latching diode, thus halving the junction capacitance as well as the reverse voltage across each diode. Not sure if it will work correctly for its primary purpose, i.e. anti-latching.
You can buy this one.
Damir
Ultra Low Distortion 0 00002 1kHz Sine Generator Assembled and Tested PCB | eBay
I bought Viktor's first unit. Fine THD but noise level is higher than I'd like and very limited in frequency selection. Its the noise level now which gets in the way of seeing the harmonics most of the time. The THD+N number (without FFT averaging to remove noise) is what i am looking at. I did a simple mod on a KH4402B ultra low distortion generator and it has selectable freqs.... now typical 1KHz thd+N is .00015%. others are better but noise is limiting them. Looking for an opamp-like circuit that can be used in other ultra-low thd+n oscillators. So far the ShibaSoku generator has the lowest noise by 10 dB and as good as Victors THD. I am up against an analyzer noise floor around -150-160 re 1v. So there is room to go lower in thd plus noise (THD+N) from generators/sources.
Thx-RNMarsh
Last edited:
yes, all the time. I am attempting upgrading the analyzer for lower output noise from its residual monitor output at this time to see if this helps. Another -6dB even would be useful to me. Thx-RNMarsh
RNM
You did not answer the questions asked by Dimitri.
The way I see it:
You need to be more specific.
The first thing to do is to set up a complete specification, so there is something to work out from.
It is certainly possible to design a circuit with THD + N in the-150dB to-160dB region, but it depends on your specifications.
BTW: with specifications in the area which it seems like you're talking about, you can forget about any circuit that resembles the SW-OPA, it is not good enough neither when it comes THD nor N.
S
You did not answer the questions asked by Dimitri.
The way I see it:
You need to be more specific.
The first thing to do is to set up a complete specification, so there is something to work out from.
It is certainly possible to design a circuit with THD + N in the-150dB to-160dB region, but it depends on your specifications.
BTW: with specifications in the area which it seems like you're talking about, you can forget about any circuit that resembles the SW-OPA, it is not good enough neither when it comes THD nor N.
S
Thanks, great idea and works even in rudimentary form - bootstrapped a single diode-drop below the emitter of the emitter follower. A dual-diode bootstrapped from the emitter of the VAS may do better - I'll check it later.
The simulation schematic below gets unity-gain THD20 down to about -135 dBr at an output amplitude of 4V into 600R - within shouting distance of -140 dB. Biasing the output stage to Class-A and a few other tweaks will probably help close the gap.
Unfortunately, I didn't squeeze the LTP Wilson CM into the board I'm expecting back shortly. However, the bootstrapping mod requires just two parts, and can probably be kludged in by piggy-backing the diode and resistor with some super-glue, etc.
Edit: At an output amplitude of 1V into 600R, H2 comes in at an even -150 dB, exceeding RNM's requirement comfortably, even with the Class-AB output stage biased at 3.5 mA.
Attachments
Last edited:
The reason I know the 990 style anti-latch diodes in degrade performance is I've tried it in real life on other stuff
I'm no SPICE guru so can't tell you if BAV99.txt correctly models junction capacitance modulation.
First guy to try a BAV99 in a similar (1ppzillion THD) circuit for real and reports the results compared to collector/base of a high voltage, low Ccb driver trans. gets to go diving on the Great Barrier Reef in Cooktown with me.
But if you want a REALLY good diode in that position, use the collector/base junction of the generic pnp or npn in LTSpice. But I can't seem to find these at Digikey, Mouser or even eBay
_______________
Have you tried this for real? Bootstraps are wonky on overload ... which is why we have the anti-latch diode in the first place. Can you post a sketch or preferably a LTSpice model?jcx said:
Linux, could you do the same with your latest incarnation? I'm not sure how you two have applied the bootstrapping.
yes I have working hardware, extensive sims, Popov, Circle critera plots...
properly chosen bootstrap can be high Z and still be effective as bootstrap at audio for diode reverse charateristics but "goes away" on clipping - leaving the diodes to do their thing locally to prevent excessive windup
properly chosen bootstrap can be high Z and still be effective as bootstrap at audio for diode reverse charateristics but "goes away" on clipping - leaving the diodes to do their thing locally to prevent excessive windup
Thanks for this Linux. Can you post the actual LTspice ASC file too please. And also the models if they aren't standard LTSpice.
jcx, have you've got something similar or even a sketch of such an application?
I'd appreciate a simple explanation of how D1,2,3 act as anti-latch and how their junction cap is bootstrapped. If BAT54 are Schottkys, their leakage will be large and even worse ... variable in the long term.
Any SPICE gurus who can comment about whether BAV99.txt models junction cap modulation properly?
____________________
Another thought is that a voltage follower with +/- 15V rails is very unlikely to be overloaded so you could do without the anti-latch.
But I'm not sure how useful a low distortion follower is in REAL life. I think in most applications, you'd have gain so overload behaviour becomes important.
jcx, have you've got something similar or even a sketch of such an application?
I'd appreciate a simple explanation of how D1,2,3 act as anti-latch and how their junction cap is bootstrapped. If BAT54 are Schottkys, their leakage will be large and even worse ... variable in the long term.
Any SPICE gurus who can comment about whether BAV99.txt models junction cap modulation properly?
____________________
Another thought is that a voltage follower with +/- 15V rails is very unlikely to be overloaded so you could do without the anti-latch.
But I'm not sure how useful a low distortion follower is in REAL life. I think in most applications, you'd have gain so overload behaviour becomes important.
Last edited:
My use for a super low thd+n amplifier is in both source generators/oscillators and distortion analyzers. the existing models from various mfr'ers do not use the latest and greatest in opamps etc. Just replacing them has improved instrument performance by 20 dB or more.
The topologies are quiet capable when used with the lowest thd+n amp circuits.
At the moment I am upgrading the performance of a ShibaSoku AD725D analyzer which I can see (withOut FFT averaging out the noise) -150dB noise floor. this is seen thru the residual distortion monitor port. I think this can be improved with amp upgrades.
Now about that noise spec. Use 20KHz BW and 1KHz signal to start. But, I often use 30Khz and 100KHz BW as well.
Thx-RNMarsh
The topologies are quiet capable when used with the lowest thd+n amp circuits.
At the moment I am upgrading the performance of a ShibaSoku AD725D analyzer which I can see (withOut FFT averaging out the noise) -150dB noise floor. this is seen thru the residual distortion monitor port. I think this can be improved with amp upgrades.
Now about that noise spec. Use 20KHz BW and 1KHz signal to start. But, I often use 30Khz and 100KHz BW as well.
Thx-RNMarsh
Last edited:
...for my op amp multiloop circuits - I haven't simmed Scott's discrete front end
looking back over the thread I see that ultra low leakage was seen as desirable for the D3 diode - Pease article warned that leakage tempco caused huge increase in zero bias conductance - why I speculated that additional measures like bootstrapping series diode may be useful
but I don't have anything but handwaving for Scott's application
may be less important with "audio" mostly being "shirt sleeves" environment, discrete separates output Q heat from rest of circuit - not like monolithic op amp speced for industrial/automotive use
I've attached the .asc file - everything is stock LTspice library stuff, except the BAV99. In this case, they're all forward biased only, so the stock 1N4148 model should give nearly identical results in place of the BAV99s.
I don't know the specifics of jcx's implementation, only the general principle.
It's not necessary to limit the reverse leakage or charge-current into the junction capacitance of both D1 and D2. Only the one facing the base of the voltage follower Q1 needs to be bootstrapped - it's the non-linear current diverted away from the base of Q1 and into D2 that contributes to distortion. The other one (D1) can have relatively high charge/discharge swings without too much impact, as long as it's charged from a relatively lower-Z source (here that comes from the emitter of Q1 through the bootstrap diode D3).
So we need D2 to have a close to constant, preferably small, reverse-bias in normal operation. This ensures that there won't be much incremental current going into the junction capacitance. A small reverse bias also ensures that the DC leakage current through D2 will be small, thus avoiding DC loading/imbalance of the LTP stage. It also makes the the task of the clamp (D1) easier when necessary.
The potential for the bootstrap has to be drawn from some node which is at a similar AC potential to the base of Q1, but which *does not load* the base of Q1. The emitter of Q1 is a natural choice, and so also the emitter of Q5. The +Vcc rail is also a possibility, but it is a low-Z node, so it's not that useful. A voltage diff has to be generated for the reverse bias across D2, either capacitively or with a voltage reference. Here, forward-biased D3 acts as that voltage reference. R12 ensures that D3 stays forward-biased in normal operation. R12 could be replaced by a JFET CCS if there's a compelling need to keep D3 at a very constant forward-bias.
During overload/clipping, we want D2 to become forward-biased, which happens when the combined potential across D2 + D1 turns sufficiently positive. It shouldn't be too high, i.e. more than 2 Vbe, or Q5 will go into saturation without the clamping action of D1 + D2. Hence the need for at least one Schottky in series in D1 and D2. I tried it with two Schottkys, and it still shows acceptably low charge/discharge currents into D2, so I left it unchanged. D2 could be a standard or low-leakage Silicon like a BAV99 also, in which case Q5 will go further into saturation before clamping begins. If both are Schottkys, it probably helps to go with a pair with relatively high reverse-breakdown voltage (maybe a bit higher than the 30V specified for the BAT54).
During normal operation, the AC impedance to ground at the junction of D1 and D2 includes (R10 || R12), which is fairly low - around 2k as shown. However, during clamping, D3 switches off and D1 is looking only at R12 to AC ground, which is relatively high-Z (100k) so the clamping action of D1 + D2 is relatively unhindered (as jcx mentioned in post #2811).
Upon reflection, an NPN emitter-follower probably works better than D3 as the bootstrap reference. A diode in series with the emitter may be even better.
Attachments
Thanks for your ASC file & bootstrapping explanations, linux.
As in this case, you need to avoid having the diodes affect distortion.
It's not just power-up which is critical. Overload behaviour (especially on LF filter circuits) and phase reversal are nasty too. More audible than 1ppzillion THD.
If you could post a schematic which shows the bootstrapping, it would be much appreciated.jcx said:
You'll note SW-OPA needs input clamping diodes AND a diode or two elsewhere to avoid latching.RNMarsh said:
As in this case, you need to avoid having the diodes affect distortion.
It's not just power-up which is critical. Overload behaviour (especially on LF filter circuits) and phase reversal are nasty too. More audible than 1ppzillion THD.
Keep in mind that an FFT is being used.
Therefore, the band width is 1Hz or less per bin.