You probably saw my comment that Richard Lee's BJT OIC proposal had a Spice ULGF of 13 MHz (and PM < 60) so I am a bit sceptical
But much of the compensation is done by the poles from transistor capacitances and roll offs.
He says he did lots of work to tweak it(and perhaps had some luck) but the result is very neat.
Need to study it to work out what controls the poles and zeros.
Hi David,
Yes, I've seen it and I'm skeptical too, but I'm even more skeptical about my own sim (with the verticals). Also in the case, Cgd is part of the compensation. However, Cgd is highly dependent of Vds. At low output levels it might work, but when the output approaches the supply voltage, the phase margin drops to an unacceptable low level. So forget that sim. Although vertical MOSFETs are very fast, the variable Cgd spoils a lot of things. Is this respect are laterals easier to handle, though gm is rather low and gm/$ very low.
Oh, you mean that thread! No wonder I've missed it.
Cheers, E.
I didnt provide the schematic I provided simplified schematic only, there are some details missing and the product doesnt use an output inductor either. I used it as not to show how stability is achieved into any speaker load without it. Seems that old cyrus 3 were thinking along the same lines. There is also a discrete servo mechanism used.
I have actually simmed the cyrus 3, it shows ULGF above 2 MHZ so I dont know where you got your figures from. But try some later models, I will not provide schematics though.
Push pull CFA topology amplifiers were in production by Pioneer in 1978, the MZ1. It was soon followed by Harmon Kardon XX who most probably copied the idea as the design is very much the same. Pioneer filed patents but these were delayed for 5 years from their filing. Harmon Kardon fans claim it is the best amp ever produced by the company and second hand examples go for more money than some expensive so called modern high end amps.
High manso,
And neither do I. Without knowing the values of C119/120 and R139/149 you can't make any prediction about the ULGF, and solely based on the value of C107/108 (=150pF). Anyhow, 3MHz is certainly feasible.
Cheers, E.
Fair in argument but far from ideal, parasitics from flaws such as in layout and component choice play an important role, as does zobel network or any stabilizing network one might use at output.
Anyway on the latest interation of the amp a different techinique is used. It is on par performance wise and Ill show the method applied to the old version of the amp as no one has as yet. Basically one ends up with a two pole response but using shunts. See the results, youll notice how the response looks very much like that of two pole miller based comp.
Performance Old version 3.18 ppm 20khz 50 watt
Latest 3.21 ppm 20khz 50 watt
During my short holidays I tried the alexander comp, and its not feasible on my design and I havent had time to work out exactly why. The result was oscillation near clipping levels. Sim shows workable margins but do display slight peaking in closed loop. TPC Im yet to experiment with but I doudbt it will result in better performance as achieved with the more recent technique we use.
Attachments
Hi Edmond
During its production life there were changes made although it was originallly released with only C107/108. I simmed that amp some years ago and if i remember correctly ULGF was a fair tad above 2 MHZ.
The author of the link you posted had an interesting CFB design with very high bandwith as well but I notice it was removed. Pity.
I would be very curious about how manso simulated the cyrus amp, as none of the critical active devices (input stage, VAS/TIS, output stage) seem to have spice models. However, one doesn't need any simulation to estimate the unity (open loop) gain frequency, and then the ULGF.
A peek at the 2SA1038 and 2SC2389 ROHM data sheets will reveal the rather low Early voltage of the 2SA1038 device: anywhere between 100 and 150V.
Ignore for the moment C119/C120 effect (they certainly won't increase the bandwidth) and look at the VAS only. Assuming the 2SA1038 is controlling the impedance at the VAS output (that is, ignoring the 2SC2389 output impedance, the output stage input impedance and the output stage input capacitance, all very rough approximations, but definitely having a bandwidth lowering effect), the C107 and C108 (to a total of 300pF) see an impedance of VA/Ic = 150V/3mA = 50k.
Something which is called in EE the short/open time constant method tells that the bandwidth is defined by the sum of all time constants in the circuit, calculated individually, while the other are short/open. As we ignored anything else except the VAS output impedance and the 300pF capacitance, these are rendering a time constant of 50k*300pF=15uS
Further, we need the open loop gain. Very hard to estimate without having an output stage model, but assume again an extremely conservative model that considers the TIS/VAS load defined by the TIS/VAS output impedance. So we have the input stage gain of R111/R109 multiplied by the TIS/VAS gain of gm*VA/Ic~40*VA=6000 (the effect of the TIS/VAS degeneration was also ignored, it actually lowers the gain). This make for an open loop gain of 240,000 or about 107dB. Subtract the closed loop gain (34dB) and you'll get a LF loop gain of 73dB. Combined with the open loop bandwidth, this renders an ULGF of 2.5MHz.
Now recap the approximations we did:
- Ignored the C319/C320 effect (lowering the bandwidth)
- Ignored the output stage input impedance (lowering the open loop gain)
- Ignored the output stage input capacitance (lowering the bandwidth)
- Ignored the TIS/VAS degeneration (lowering the open loop gain)
- Ignored the output impedance of the other TIS/VAS half (lowering the bandwidth and the open loop gain).
If the devices models would be available and a more exact determination would be possible, then make no mistake - the ULGF will be much lower than the above very rough calculation. Hence my estimation of 1MHz.
Now you know where I got the number. I can't hope you'll appreciate the lecture, though.
Last edited:
And because I was at it, I did a quick simulation, using the best devices I had models from (2SA970/2SC2240 in the input stage, 2SA1381/2SC3503 (with Early voltages close to 800V) in the VAS, same as pre-drivers, 2SC5171/2SA1930 (with Ft=200 MHz) as drivers, MJL4281/MJL4302 (with Ft=30MHz) as outputs), the results are attached.
As expected, the ULGF is 1.25MHz (compared to 2.5MHz above) and it looks very stable. LF loop gain is 65dB (compared to 73dB estimated above). I have no doubts that the inferior devices used in the original design (to mention only the output devices with Ft=20MHz and the VAS devices with 150V Early voltage) will bring the ULGF at or under 1MHz, and this is even before considering the effect of C119/C120 (of unknown value).
Bottom line, this Cyrus CFA has an ULGF at or under 1MHz (which is typical for commercial designs). I am still waiting (but not holding my breath) for an example of commercial amp with an ULGF of 3MHz, 6MHz or above.
As expected, the ULGF is 1.25MHz (compared to 2.5MHz above) and it looks very stable. LF loop gain is 65dB (compared to 73dB estimated above). I have no doubts that the inferior devices used in the original design (to mention only the output devices with Ft=20MHz and the VAS devices with 150V Early voltage) will bring the ULGF at or under 1MHz, and this is even before considering the effect of C119/C120 (of unknown value).
Bottom line, this Cyrus CFA has an ULGF at or under 1MHz (which is typical for commercial designs). I am still waiting (but not holding my breath) for an example of commercial amp with an ULGF of 3MHz, 6MHz or above.
Attachments
Last edited:
There are two important points that bear.
Firstly, the SMALLER the loop, the more stable its likely to be. eg a 2nd order loop is theoretically ALWAYS stable. (apologies to Waly for ignoring his esoteric cases
) If it is unstable, you have turned it into at least a 2nd order loop .. probably cos evil parasitics. A smaller, simpler loop is also less likely to have evil parasitics cos your layout can be simpler & tighter .. and hence will be stable with a higher ULGF.Secondly, and related to the above is Cherry's approach which I paraphrase in #163 tpc-vs-tmc-vs-pure-cherry
BTW, Cherry's ETI apr83 article is on Jan.didden's website
A Locanthi triple, QUAD triple, 'EF2+single VAS' (all 'Inner Loops' including the Locanthi triple EF) with MJ3281/1302 all have a Current gain ULGF much higher than 30MHz.
But Current gain, though appropriate to work out how much they contribute to the overall Loop Gain, isn't the right metric to judge the stability of the 'Inner Loops'.
The key to stability in all 3 cases is reducing the Gm of the 'inner loop'.
The probe at the IPS shows stability of the whole amp. You need to look at both.
At the risk of generalising, IMHO, the IPS probe is the more important one. The 'inner loop' is just one of the blocks within the 'main loop'. The 'main loop' inherits all the evil very HF poles of the 'inner loop' but if these are relatively even higher for the 'main loop', they are innocuous.
For the .ASC I posted in #1480, reducing C1/3 reduces LG for the 'inner' Cherry loop (ie makes it more stable) but increases it for the main 'IPS' loop. This reduces the stability of the complete amp and you can check this by taking it to extremes and using the other 'non-linear' measures like various capacitive loads and .TRANS
This is a valid technique but no more useful than increasing one of the 'more' dominant poles within the loop or trying various capacitive loads. IMHO, the last is the most useful cos it ties in with important 'real life' tests which you need to do.
In Is Linear Phase Worthwhile?, I show the 'real life' delays due to simple 1st order poles and also point out where the concept of 'group delay' is appropriate or not.
some of which I discuss in the tpc vs bla bla link above.
If this is astx's supa amp, you'll remember we couldn't sim his stability results either.May I mention that the one time we have both analysed and made our respective recommendations and then had them checked on the same real amplifier that "Pure Cherry" came out as less stable and lower performance
The closest I got was simulating his 4" of ribbon cable but even then it was 'far' from real life though we moved substantially closer on other points due to your expertise as a SPICE guru
My take on that very educational project was that to go further, we needed to do a LOT more work to move the SPICE model closer to 'real life' .. or preferably make another 'real life' version with even tighter layout bla bla eg definitely without the ribbon cable .. ie move 'real life' closer to SPICE world
Thanks for your kind words.
I think the point about making the 'inner Cherry loop' more stable reduces the stability of the complete amp is the relevant one. It is borne out in 'real life'.
Thanks for these Edmond and the links. They'll allow me to sim more Jurassic 'real life' experience.
But how do you invoke them?
Is it
2sj201 or 2sj201_DC,
2sk1530-Y or 2sk1530_DC ?
Excuse da newby questions ..
Last edited:
Maybe cos our simple Jurassic common techniques give better slew/THD/sound/bla bla than supa dupa uncommon methods.
I'm sorry these old,naive common methods were not invented & built by virgins from liquid Unobtainium
If your supa dupa complex uncommon methods give better results, we would be
pleased to hear from your virgins.
Maybe cos our simple Jurassic common techniques give better slew/THD/sound/bla bla than supa dupa uncommon methods.
I'm sorry these old,naive common methods were not invented & built by virgins from liquid Unobtainium
If your supa dupa complex uncommon methods give better results, we would be
pleased to hear from your virgins.
I am 13yrs designing building and mod-ing audio amplifiers. May be not as old as you guys here that has decades of time, anyway, just use the common, but do not too much dream.
Edit:Complexity and speed superiority won't help in sound quality. There is another matters, that easier to manage with complexity.
Last edited:
The Cyrus does not have an OP inductor in your sim Waly ( is that the case with the production version?). It's not going to handle capacitive loads very well, and especially so in light of the OPS triple.
I am also interested at the R22 and C3 network across the FB resistor.
I would have comp'd this amplifier differently.
I am also interested at the R22 and C3 network across the FB resistor.
I would have comp'd this amplifier differently.
Last edited:
Wanna post one of YOUR designs with performance us old fogeys can only dream off?I am 13yrs designing building and mod-ing audio amplifiers. May be not as old as you guys here that has decades of time, anyway, just use the common, but do not too much dream.
Us oldies are never too old to learn from true gurus
State of Art?
This has been a very educational thread.
It has exploded many of the lies that evil VFA acolytes have been spreading about CFAs, poor THD, poor noise etc.
We now know CFAs can better VFAs on noise and push them very hard on THD .. all with very simple circuits which IMHO, is their main advantage along with better sound of course.
But alot of what we have been solving isn't about CFAs but about 'symmetrical' amps in general.
We have heard very little about CFA IPS and no sensible discussion.
I've proposed the 'simple CFA' like VSSA and my #500 and pointed out that in most cases, this will usually give up to 3dB better Loop Gain and hence THD .. as well as being simpler. The disadvantage is evil electrolytics.
Two different Diamond Inputs have appeared but with no discussion about their advantages.
Playing with Bonsai's excellent nx-Amp suggests to me that one of the disadvantages of 'simple CFA', poor control of VAS current, is also a problem with Diamonds.
The two Diamond variations are
- Dadod's : where the Diamond has 2 Baxandall pairs such that the input stage is bootstrapped
and
- Bonsai's : with input collectors taken to a low voltage rail.
Anyone willing to pontificate on these?
There was a Studer/Revox version without emitter resistors in the input Diamond devices. But having played with this, I'm consigning it to 'hand carved by virgins' territory. DC conditions are too twitchy.
____________________
Lastly, I hope Guru Bob Cordell is lurking and I sincerely hope he's encouraged to do more than damn CFAs with faint praise in the next edition of his book.
Maybe even do some 'real life' work on the subject.
This has been a very educational thread.
It has exploded many of the lies that evil VFA acolytes have been spreading about CFAs, poor THD, poor noise etc.
We now know CFAs can better VFAs on noise and push them very hard on THD .. all with very simple circuits which IMHO, is their main advantage along with better sound of course
.But alot of what we have been solving isn't about CFAs but about 'symmetrical' amps in general.
We have heard very little about CFA IPS and no sensible discussion.
I've proposed the 'simple CFA' like VSSA and my #500 and pointed out that in most cases, this will usually give up to 3dB better Loop Gain and hence THD .. as well as being simpler. The disadvantage is evil electrolytics.
Two different Diamond Inputs have appeared but with no discussion about their advantages.
Playing with Bonsai's excellent nx-Amp suggests to me that one of the disadvantages of 'simple CFA', poor control of VAS current, is also a problem with Diamonds.
The two Diamond variations are
- Dadod's : where the Diamond has 2 Baxandall pairs such that the input stage is bootstrapped
and
- Bonsai's : with input collectors taken to a low voltage rail.
Anyone willing to pontificate on these?
There was a Studer/Revox version without emitter resistors in the input Diamond devices. But having played with this, I'm consigning it to 'hand carved by virgins' territory. DC conditions are too twitchy.
____________________
Lastly, I hope Guru Bob Cordell is lurking and I sincerely hope he's encouraged to do more than damn CFAs with faint praise in the next edition of his book.
Maybe even do some 'real life' work on the subject.
EKV models
Hi Richard,
It is 2SJ201 and 2SK1530-Y. These are sub-circuits you should invoke. (the sub-circuits in turn invoke the xxxx_DC models)
In the file below I have changed 2SJ201 into 2SJ201-Y, which is a more appropriate name when vto= -1.72 (the gate threshold voltage).
Good luck with this stuff.
Cheers, E.
PS: How to invoke the sub-circuits in LTSpice, I don't know exactly, as I'm a MiroCap user.
Most likely, you have to create new components first, which in turn refer to these sub-circuits.
Hi Richard,
It is 2SJ201 and 2SK1530-Y. These are sub-circuits you should invoke. (the sub-circuits in turn invoke the xxxx_DC models)
In the file below I have changed 2SJ201 into 2SJ201-Y, which is a more appropriate name when vto= -1.72 (the gate threshold voltage).
Good luck with this stuff.
Cheers, E.
PS: How to invoke the sub-circuits in LTSpice, I don't know exactly, as I'm a MiroCap user.
Most likely, you have to create new components first, which in turn refer to these sub-circuits.
Hi forr,
That's right. It were simple amps with feedback to the emitter of the input stage which comprises a single tranny and were sometimes preceded by a jfet.
Later on, in the seventies, I switched to LTP input stages.
In the fifties I also built a few (low power) transistorized amps with small transformers in the class-AB OPS*, which was common practice in those days.
Cheers, E.
* BTW, I once owned a MC2300 with huge OP transformers. It was ridiculous heavy amp: 128 pounds!
No its not, I wanted to answer your post yesterday but I prefer my wife to your ramblings and it was a bit late. Just to rekindle my memory I simmed the amp as well, as expected the ULGF was 2.4 Mhz, actually very close to 2.5MHZ. Your estimation is feasible but there were just too many assumptions which makes it too inaccurate.
Maybe you dont have models for the transistors used but I do and yes they are available. Im will sell them to you if you interested. They are not cheap at all but I as user and the company that produced them can gaurantee their conformity to the highest level. You just need to supply the batch production number for best model matching.
As I dont trust simulation that much I repeated using a different simulation program, Multisim. ULGF 2.3 Mhz. Not much in it compaired to LT.
I wonder how you can be as far off as 1.25 Mhz ??
Give me a day or two and Ill post my results.