Great, we are now thinking about the current-mirror; I think I suggested the cascode already. Here's some comparisons using bjt --
Thx-RNMarsh
Thx-RNMarsh
Last edited:
Do you have the #?
Frankly I don't think it meets the test of sufficient novelty, but as it is held by a big company I suppose one should tread with some caution. I wonder if they cited Boxall or Larson?
I have the EU patent registration, not sure why it needed a EU patent besides a US one. I ll post number tomorrow when Im at work.
I was near to using it till I found the patent. I dont think texas will let anyone off the hook if used commercially, big company means many many lawyers. 😱😀
What sometimes works, if there is a real concern, is to point out that you are using previous expired patents, and/or obviously prior art. In this case running a wire between Boxall's patented (long, long ago expired) circuit and a standard e follower enhanced current mirror (probably never patented) is surely all that you are doing. Use a small resistor in series for good measure.
But TI probably couldn't care less about, say, someone's "boutique" power amp topology. They make ICs and market them aggressively, as well they should. They are about as likely to get into high-end audio as I am to be picked to be goalie for Argentina in the next World Cup.
I see no comments in the table about linearity Richard, whic in our field of interest would be critical.
Clearly, a straight Wilson has some accuracy problems, and in a balanced topology amp that might lead to uneccessary offsets before any servoing is applied. In the LTspice file I posted up, I have a few of these types.
To be honest, I was not able to see great advantages over the simple two transistor mirror with helper transistor - the performance looks remarkably good for the parts investment. Of course, the output impedance does not match the more advanced designs, so there may well be a linearity issue right there.
Brad, I'll have to take a look at your current conveyor circuit in the next few days. Matching scares me though!
Last edited:
Do these speakers (yours & the e-JMLC) have active xovers and a separate amp driving the treble?
What size is the box on the e-JMLC? What bass cutoff and sensitivity?
____________________
Your system will indeed never clip if you don't touch the volume control. But more than 60% of Pano's little poll will play louder, some 28% MUCH louder than you.
And it will not come anywhere near playing 'The Garage Door' at the correct level. 🙂
Really !?
See the column labelled "Accuracy".
"For high performance analog circuit apps, the accuracy and output impedance are the most important parameters to determine the performance of the current mirror."
[Electronics, Circuits and Systems, 2001. The 8th International Conference on (Vol 2. pgs 565-568)]
Thx-RNMarsh
Last edited:
I agree with what you have said, and you made multiple points. First, in general, if something must clip, don't let it be the power amplifier-proper. Do the clipping in a line-level small-signal location. Make it as sharp or as soft as you like, to your taste and priorities.
Amplifiers in the 99% of consumer audio, including hi-end, where clipping is essentially uncontrolled (i.e., no compressors, limiters, clippers, etc in the signal path), amplifiers do indeed clip when playing well-recorded program material at realistic SPL levels with typical loudspeaker sensitivities. It would be nice if consumer audio amplifier designers would put clip indicators on their amplifiers. That would be enlightening to many. My favorite for crest factor is Rickie Lee Jone's "Ghetto of My Mind" on her Flying Cowboys album.
Although the program material on most vinyl does not warrant huge overload margins, ticks and pops can create quite high output levels from cartridges. These interferences may be less audibly annoying if they are clipped cleanly. The phono preamp guy just doesn't want that clipping to occur in his part of the signal chain. It probably also makes a difference where in the RIAA equalization process the clipping occurs if it is in the phono preamp.
Cheers,
Bob
Bob, I'd say given how easy it is to design in headroom in the small signal stages, in most cases you are going to have to deal with it in the power amp, and there, ultimately its the rail voltage and current delivery that sets the limit for a given load.
(Caveat: I note that there are many RIAA amplifiers that have dismal overload charateristics - John Atkinson has found a few with <10dB and I saw one IIRC that he measured with just 3dB - really unacceptable nowadays IMV since it is so easy to design to accomodate 25 or 30dB peaks in a RIAA)
(Caveat: I note that there are many RIAA amplifiers that have dismal overload charateristics - John Atkinson has found a few with <10dB and I saw one IIRC that he measured with just 3dB - really unacceptable nowadays IMV since it is so easy to design to accomodate 25 or 30dB peaks in a RIAA)
Thanks for your support Bob 🙂
Soft clipping is of course anathema to those who want nothing intruding until the drastic events 🙂 But perhaps we should devise some "eventual soft clipping" which completely stays out of the way until engagement.
Another project!
For myself.... soft clipping is fine IF you need to have it/use it. Do I need to have it? I do not. If I have at least 250W/8 my system wont be clipping in my room size and speaker effec. Now, if I used a 20W or 50W, yes then a soft clipping circuit would be helpful. But why would I use that or design that for my home system? [personnal value judgements aside - like costs too much etc]
Thx-RNMarsh
Thx-RNMarsh
Last edited:
Interstingly, CFA's tend to clip softer than high loop gain VFA's - I noticed this on my to efforts.
I've done a quick sim and am struggling to see what it brings to a straight 3 or 4 transistor mirror. Bandwidth is slightly better, but its more peaky, and distortion is not as good as the 4 transistor version (30ppm at 5mA).
The cascode certainly adds nothing in terms of bandwidth in a practical amplifier because of the capacitive loading on the cascode output will dominate.
Any thoughts?
If you look In circuit i posted, i use them as well.
If i remember right, i Got at enhanced phase margin
Typically, using the Boxall compound pair, or other control-electrode-recycling scheme, yields a reduction in output capacitance (until the transistors poop out at VHF), and hence a reduction in the variation of that capacitance with voltage, in turn thus a reduction in HF distortion. If the external capacitive loading is mainly constant with voltage you retain the distortion advantage. And one of the primary ways to compensate the circuit's peaking is adding a little output C. Also, if the input current comes from a very low capacitance source, a little lumped input C at the mirror input helps.
I haven't simulated the exact circuit shown as yet. But one thing about Boxall which Walt J. and others have noted, is that, to work well it really needs to be fed, as a compound common-base stage, from a relatively high source impedance. I believe this is because the returned base current divides between that impedance and the main device emitter. This is true of the Aldridge as well, and that takes up usually a greater amount of voltage.
Jaeger published as a brief mention a compound Aldridge-style cascode with a PNP and a cascode JFET. Walt discovered that Csanky disclosed the two-JFET version. All this stuff is quite old now. I will be interested to see if the TI patent mentions any of it. You would think that the USPTO could at least find old patents, even if it misses non-patented prior art 🙄
I recall the famous NAD3020i amplifier (clipping at 30 W - 8 Ohm) was provided with a switchable input soft clipping circuit. Reviewers said that switching it on was only to be recommended during parties and should be avoided for serious hi-fi listening.
I'll have to try feeding it from a cascode (which is what will happen in a typical CFA).
Re the output capacitance - I can see that between the output mirror transistor and the input to the output boxall pair you can keep capacitance low, and hence the BW potential, but my earlier point was what happens on the output of the Boxall pair? In an IC you can control capacitances quite tightly and keep them low. In a discrete design I think this will be more difficult. In the sims I added Cstray - it really does impact BW. However, it is higher than we need I'd say.
Is that DC accuracy or accuracy at say 100 kHz? it was not clear to me from the table. I will post up the results shortly.
Last edited:
Here is the mirror sim file updated with the Boxall cascode
BASIC WITH HELPER
Total Harmonic Distortion: 0.009537%
4 TRANSISTOR WILSON
Total Harmonic Distortion: 0.001217%
BASIC
Total Harmonic Distortion: 0.497355%
BOXALL CASCODE
Total Harmonic Distortion: 0.009121%
I've attached the responses in the pdf - the traces correspond to the .asc Vo numbers.
BASIC WITH HELPER
Total Harmonic Distortion: 0.009537%
4 TRANSISTOR WILSON
Total Harmonic Distortion: 0.001217%
BASIC
Total Harmonic Distortion: 0.497355%
BOXALL CASCODE
Total Harmonic Distortion: 0.009121%
I've attached the responses in the pdf - the traces correspond to the .asc Vo numbers.
Last edited:
I went and did some sims as well. To dramatize the advantages of Boxall added to e-follower enhanced mirror, due to the output capacitance reduction and consequent distortion from voltage-variable capacitance, the setup was 200 ohm emitter ballasting, 5mA quiescent, only +/- 40uA signal current (to make the nonlinearities of the basic mirror small). The load R to a 12V reference was 100k. Frequency 10kHz. Boxall auxiliary Q biased at 200uA, its base at 3.4V (this roughly equalizes the dissipations and their signal-induced shifts in the two mirror transistors).
So with the output voltage swings of ~8V p-p, the unadorned mirror with harmonics to 15th: 1703ppm. With Boxall, 4.74ppm. The C load for Boxall for flat response is about 0.4pF, of course unrealistic, but with that a -3dBr frequency of about 5MHz. The distortion is unchanged with 5pF, bandwidth drops to 319kHz. The simple mirror bandwidth is about 412kHz with no external C loading. Transistors are 2SC1815 and 2SA1015. The E followers have 3k to common.
The auxiliary transistor can be very fast and small geometry. John Addis told me of a current source he made for an instrument that used an obsolete 🙁 Philips part for the main device, I think about a 1GHz high voltage one optimized for CRT deflection, and a ~5GHz one for the auxiliary. Sims and builds revealed a nasty parasitic oscillation due to emitter inductance in the main device, and it was successfully quelled with an R-C network in series with the emitter, small enough that it didn't interfere otherwise with circuit operation. The I source had a large voltage compliance and a sub-1pF output capacitance out to rather high frequencies 😎
So with the output voltage swings of ~8V p-p, the unadorned mirror with harmonics to 15th: 1703ppm. With Boxall, 4.74ppm. The C load for Boxall for flat response is about 0.4pF, of course unrealistic, but with that a -3dBr frequency of about 5MHz. The distortion is unchanged with 5pF, bandwidth drops to 319kHz. The simple mirror bandwidth is about 412kHz with no external C loading. Transistors are 2SC1815 and 2SA1015. The E followers have 3k to common.
The auxiliary transistor can be very fast and small geometry. John Addis told me of a current source he made for an instrument that used an obsolete 🙁 Philips part for the main device, I think about a 1GHz high voltage one optimized for CRT deflection, and a ~5GHz one for the auxiliary. Sims and builds revealed a nasty parasitic oscillation due to emitter inductance in the main device, and it was successfully quelled with an R-C network in series with the emitter, small enough that it didn't interfere otherwise with circuit operation. The I source had a large voltage compliance and a sub-1pF output capacitance out to rather high frequencies 😎