Bob Cordell Interview: BJT vs. MOSFET

Jump to Latest
#1,441
2 to 10 uf should be OK, because many electrostatic loudspeakers have this capacitance. I usually show the JC-1 with big electrostatics at CES.
However, Bob is trying to paint me into a corner by specifying multiple 2 uF ceramic caps (hardly a realistic load) just to challenge me.
It would be like if this were an auto website, and different auto designers made their cars somewhat differently. Let us say that some guy made limos for diplomats and knew he had to make it safe in a getaway from terrorists. He might specify that the car clear a speed bump at 100mph. Then he might challenge the Porsche designer, IF he can do the same. Maybe, the Porsche can, and maybe it can't, but it would not be the first thing that the designer would test for, and a lower car might be a better performing car.
This ceramic cap thing is much that same sort of challenge.
 
#1,442
I assume the uF # is from the ? nF es loudspeaker cap reflected at the transformer primary

I expect that the es loudspeaker xmfr has noticable series R and very finite bandwidth - there should be no need to drive low esr uF at typical ss amplifier loop gain crossover frequencies

anyone have some #s?
 
#1,443
Amplifier Load Isolation... or Not

Bob Cordell said:


Obviously, the coil was needed in the past. How do you get away with not using it now? For example, an answer might be that you chose to use low amounts of feedback. Or, that by using a huge number of fast output devices, you have shown it to be unnecessary. Or that you have used a proprietary form of feedback compensation. That's what I was getting at.

It is exactly the question of an amplifier driving an indeterminate load that prompted me to investigate an EC configuration, in which there is no global feedback. I took the opportunity to review the Hawksford and Cordell EC articles as a starting point. In looking at the EC output topology, it occurred to me that a single error correcting point, if feasible, might yield better than the ~20 dB of error correction stated by Cordell in his 1984 AES article. In a previous thread I showed simulation results of an ideal source driving an EC stage utilizing a high BW, high current video opamp with its +/- 15V supplies floating on the output signal. The results show distortion suppression in the range of ~55 dB at 300 Hz to ~20 dB at 30 KHz. These results were sufficiently encouraging that I took the next step and coded up an entire amplifier design.

As a next step I simulated the entire amplifier (no ideal sources, level 3 HSPICE models) with the video amp EC back end. Feedback was employed between the input stage and the follower stage that drove the EC, but no GFB was used. The output load consisted of 4 ohms || 5.0 uF with no isolation network.

For the initial distortion tests, the input voltage was adjusted to yield a 100 VPP sine wave at the input of the EC amp. Distortion at the input to the EC amp hovered around -117 dB below the main signal, with an increase to -110 dB at 20 KHz. Distortion at the load remained at ~-114 dB from 20 Hz - 2.0 KHz, rising steadily to ~ -90 dB at 20 KHz. Distortion products were monitored out to 100 KHz, so some of the high frequency distortion products lie outside the audible range. Greater distortion suppression at high frequencies was not possible owing to the need to maintain EC loop stability. The amplifier remained stable for all inputs, including a -50 to + 50V step with a 0.5 us rise time at the input to the EC amp, which drew a transient current of >100 amps!

Based on these result, I have decided to pursue an EC-based design rather than a a conventional GFB design, even though the latter yields lower distortion. Needless to say, the NFB design without load isolation went into immediate oscillation with a 4 ohm || 5 uF load.

I'll post some waveforms and distortion figures when I have time to plot them.

Comments are welcome.
Click to expand...
 
#1,445
Re: Amplifier Load Isolation... or Not

analog_guy said:
...
Distortion products were monitored out to 100 KHz, so some of the high frequency distortion products lie outside the audible range.
...

Comments are welcome. [/B]
Click to expand...

In my experience these high frequency distortion products above audible range are very well audible, as they get folded down through intermodulation.
In my opinion one of the key parameters making an amp sound good/bad.

Mike
 
#1,449
john curl said:
Bob, I would not deliberately subject my power amp to paralleled 2uf ceramic caps across the output terminals, but it would be a tough test. I hope that no audiophiles out there will do it either.
Click to expand...


So it sounds like your amp is not unconditionally stable into a capacitive load. That's OK. It is a reasonable choice.

I'm not trying to paint you into a corner, I'm just trying to get the facts so we are all on a level playing field. If you think that running without any output coil at all in a home environment into electrostatics that are isolated to some extent from the amplifier by the speaker cable, that is perfectly fine, just say so. Many of us could do the same without an output coil, but maybe are being more conservative - perhaps unnecessarily so.

Have you SPICE'd the JC-1 into different kinds of capacitive and other bad loads?

Thanks,
Bob
 
#1,450
Re: Amplifier Load Isolation... or Not

analog_guy said:
Bob Cordell said:


Obviously, the coil was needed in the past. How do you get away with not using it now? For example, an answer might be that you chose to use low amounts of feedback. Or, that by using a huge number of fast output devices, you have shown it to be unnecessary. Or that you have used a proprietary form of feedback compensation. That's what I was getting at.

It is exactly the question of an amplifier driving an indeterminate load that prompted me to investigate an EC configuration, in which there is no global feedback. I took the opportunity to review the Hawksford and Cordell EC articles as a starting point. In looking at the EC output topology, it occurred to me that a single error correcting point, if feasible, might yield better than the ~20 dB of error correction stated by Cordell in his 1984 AES article. In a previous thread I showed simulation results of an ideal source driving an EC stage utilizing a high BW, high current video opamp with its +/- 15V supplies floating on the output signal. The results show distortion suppression in the range of ~55 dB at 300 Hz to ~20 dB at 30 KHz. These results were sufficiently encouraging that I took the next step and coded up an entire amplifier design.

As a next step I simulated the entire amplifier (no ideal sources, level 3 HSPICE models) with the video amp EC back end. Feedback was employed between the input stage and the follower stage that drove the EC, but no GFB was used. The output load consisted of 4 ohms || 5.0 uF with no isolation network.

For the initial distortion tests, the input voltage was adjusted to yield a 100 VPP sine wave at the input of the EC amp. Distortion at the input to the EC amp hovered around -117 dB below the main signal, with an increase to -110 dB at 20 KHz. Distortion at the load remained at ~-114 dB from 20 Hz - 2.0 KHz, rising steadily to ~ -90 dB at 20 KHz. Distortion products were monitored out to 100 KHz, so some of the high frequency distortion products lie outside the audible range. Greater distortion suppression at high frequencies was not possible owing to the need to maintain EC loop stability. The amplifier remained stable for all inputs, including a -50 to + 50V step with a 0.5 us rise time at the input to the EC amp, which drew a transient current of >100 amps!

Based on these result, I have decided to pursue an EC-based design rather than a a conventional GFB design, even though the latter yields lower distortion. Needless to say, the NFB design without load isolation went into immediate oscillation with a 4 ohm || 5 uF load.

I'll post some waveforms and distortion figures when I have time to plot them.

Comments are welcome.
Click to expand...


Sounds like some nice work and results. Just remember that a non-isolated low-ESR capacitive load can destabilize an EC circuit in some cases even when there is no global NFB.

Also keep in mind that a speaker cable can look like an unterminated transmission line, where very radical and funny things can happen to the impedance seen looking into it at frequencies where there may be reflections, like at the quarter wave length. This depends a lot on the speaker load at the other end. Some speakers look inductive above 20 kHz, and this can give rise to the unterminated line effect at HF. In other cases, this is mitigated by the resistive tweeter pad that is usually there. For these reasons, some advocate the addition of an R-C shunt Zobel network at the speaker end of the speaker cable. This can effectively far-end terminate the speaker cable in its natural impedance at HF.

Bob
Click to expand...
 
#1,453
For the record, I was in charge of computer simulation, at Friden in 1966, BEFORE SPICE was invented! How do I know? I was taking a class with Dr. Pederson in 1970, when he discussed developing SPICE as an improvement over ECAP, that proceeded it.
I might recommend his book that was derived from our class notes at the time:
'Analog Intergrated Circuits for Communication' By Donald O. Pederson and Kartikeya Mayaram, Kluwer Academic 1991.
 
#1,454
jcx said:
I assume the uF # is from the ? nF es loudspeaker cap reflected at the transformer primary

I expect that the es loudspeaker xmfr has noticable series R and very finite bandwidth - there should be no need to drive low esr uF at typical ss amplifier loop gain crossover frequencies

anyone have some #s?
Click to expand...

It's usually a series LRC network, resonant somewhere
between 10KHz and 20 KHz. I have seen as low as .5 ohm
resistance, not counting speaker cable.

😎
 
#1,457
andy_c said:
BJT vs. MOSFET anyone?
Click to expand...


How about BJT vs BJT? 🙂

A technical discussion on the merits of low fT audio power output BJT’s vs. high fT ones would be good.

Douglas Self’s “Blameless” does about 0.01% THD-20 with complementary 2MHz fT BJT’s in class B and even lower THD-20 with quasi-complementary (!) 2MHz BJT’s in class A. To put this into some kind of perspective, Bob’s 50W MOSFET amp only managed 0.02% THD-20 with the IRF vertical MOSFET’s without the error correction – and his basic topology has a lot more HF negative feedback than the miller compensated “Blameless”.

I don’t want to argue that low fT BJT are better than high fT BJT’s, but I think the low fT BJT’s get an undeserved bad rap. BJT’s with high fT’s have limitations in other areas too. For instance, the MJL21193/MJL21194 complementary BJT pair with a 4MHz fT has slightly lower Cib and Cob that the MJL3281A/MJL1302A high fT (30MHz) equivalent pair.

http://www.onsemi.com/pub/Collateral/MJL21193-D.PDF
http://www.onsemi.com/pub/Collateral/MJL3281A-D.PDF

See graph on the top of page 5.

Does any one know of any good applications literature delving into the manufacturing processes of high fT audio power BJT’s? This would make an interesting read.

Cheers,
Glen
 
#1,458
G.Kleinschmidt said:
For instance, the MJL21193/MJL21194 complementary BJT pair with a 4MHz fT has slightly lower Cib and Cob that the MJL3281A/MJL1302A high fT (30MHz) equivalent pair.

http://www.onsemi.com/pub/Collateral/MJL21193-D.PDF
http://www.onsemi.com/pub/Collateral/MJL3281A-D.PDF
Click to expand...

Hmm, something funny is going on there with the Cib curves of the 21193/21194. There are two decades of capacitance between the 10000p and 1000p labels. I think the 10000p label should be 100000p. The fT value is related to Cib by the set of formulas that begin here (see equation 11). So it would make sense that Cib for the 21193/21194 should be about 10x that of the 3281/1302 given that fT is about a factor of 10 lower for the 21193/21194.

I haven't seen any good info on fab details of the ring emitter devices but I haven't looked hardly at all either.
 
#1,459
andy_c said:


Hmm, something funny is going on there with the Cib curves of the 21193/21194. There are two decades of capacitance between the 10000p and 1000p labels. I think the 10000p label should be 100000p. The fT value is related to Cib by the set of formulas that begin here (see equation 11). So it would make sense that Cib for the 21193/21194 should be about 10x that of the 3281/1302 given that fT is about a factor of 10 lower for the 21193/21194.

I haven't seen any good info on fab details of the ring emitter devices but I haven't looked hardly at all either.
Click to expand...


G'day

You're right about Cib. It's obviously a printing error - a missing zero. The Cib value did leave me a bit confused and I'm embarassed that I didn't notice the obvious error in the graph. However, It's worth noting that the high ft devices are still hindered by high Cob, which in this specific case seems to be higher for the faster transistor.

Cheers,
Glen
 
You must log in or register to reply here.
Top Bottom