In what context?
If you follow his interviews, Cordell makes it clear that Re minimum should be set to suit the transistor AND the supply rail voltage. There is no Re maximum, you can go higher or much higher.
Lower supply rails allow lower values of Re.
I very much doubt you will get good current sharing with 5pr using 0R1 or even 0R15. 0R22 would be a reasonable starting point and you may find you get better results with 0R33 depending on the characteristics of your specific output transistors.
I used to use 0R1 Re's on some of my projects but have stopped using them altogether, even for simple single pair outputs running from modest rails. Why? Bias stability is not great and increasing to 0R22 makes a huge improvement.
Alan, why don't you build yourself a few simple designs, simplified versions of your grand project, if you like, and then you will get the experience of seeing the different effects?
I used to use 0R1 Re's on some of my projects but have stopped using them altogether, even for simple single pair outputs running from modest rails. Why? Bias stability is not great and increasing to 0R22 makes a huge improvement.
Alan, why don't you build yourself a few simple designs, simplified versions of your grand project, if you like, and then you will get the experience of seeing the different effects?
Thanks
I kind of question that also of follower straight Oliver's optimization.
I was set to do the straight 3EF until I saw your schematic. You get away from using CCS for the opposite pre drivers. Did you get that from some body else? If so, do you have the link on the theory so I can read it over? I want to understand exactly how it works.
Here you are saying you want to chuck the complete run of the boards of Mr. Cordell's complementary IPS design. I have that layout and ready to sent out. I am even temped to sent it out anyway!!! I am sure I can use a trim pot on one of the CCS and get them going. Hey, we are talking about high end stuff here. To me, it's perfectly ok to hand match and trim.
I was even thinking about asking you to spare me a board if you going to dump it. But a run of 10 is only $46. I am glad you decided to go ahead with that.
I am going to have a few different IPS/VAS. I do this as a system. I have one OPS board hopefully I don't have to change. Then I have 3 or 4 IPS/VAS boards. I am planning to make all of them to have mounting holes and the common signals( rails, ground, input) at the same location so I can stack them up. Then use a rotary switch to pick one output to drive the OPS!!! I am going to end up with only one amp, but I do try different IPS/VAS.
Chassis, PT and reservoir caps are the expensive part, I am not going to replicate these. I just define the mechanical mounting hole locations. So if I need new runs, I can just make new boards with the same mounting holes to the heatsink and chassis.
Thanks
Last edited:
Hi Alan
I plan to build up CSH6 (full comp) at some stage but I'm a lot less than enthusiastic about it now with the diff shunt caps @ 4k7 than I was with them at 22k. These added performance vs. simplistic resistor loads (e.g. Leach amp) just doesn't seem to justify the significant added complexity. I'm going to adapt Cordell's Figure 7.15 and use that as the basis for my next push-pull VAS project.
Considering your targets, I think you should take a long, hard look at OStrippers 5pr output board, designed for modular systems that allow the front end to be swapped in and out. I bought a bunch of these boards myself and they're great. I don't think you will do better than this.
The EF3 on my CSH6 uses a diamond buffer driver stage, with the pre-drivers' current sources bootstrapped to the output node to improve rail efficiency on this low power design. The complex front-ends you are considering will eat into your precious Vrails (only 40V for 5pr!!) so you will probably want to add a low current sub rail to run the front-end and driver stage.
I plan to build up CSH6 (full comp) at some stage but I'm a lot less than enthusiastic about it now with the diff shunt caps @ 4k7 than I was with them at 22k. These added performance vs. simplistic resistor loads (e.g. Leach amp) just doesn't seem to justify the significant added complexity. I'm going to adapt Cordell's Figure 7.15 and use that as the basis for my next push-pull VAS project.
Considering your targets, I think you should take a long, hard look at OStrippers 5pr output board, designed for modular systems that allow the front end to be swapped in and out. I bought a bunch of these boards myself and they're great. I don't think you will do better than this.
The EF3 on my CSH6 uses a diamond buffer driver stage, with the pre-drivers' current sources bootstrapped to the output node to improve rail efficiency on this low power design. The complex front-ends you are considering will eat into your precious Vrails (only 40V for 5pr!!) so you will probably want to add a low current sub rail to run the front-end and driver stage.
I meant to say to you that with trim pot on the CCS, I think you can stay your course of 22K or higher. I won't do 4.7K as you said, there is no more advantage over the Leach amp anymore. I truly believe you can keep the 22K. If I were to do it, I'll keep the 47K and work with it. I found a matched transistor for the differential pair. With a trim pot at the CCS, I think I can do it.
I really don't care to conserve the peak voltage output. I always plan of 30Vpeak to get 0.5 X ( 30V X 7.5A)=112.5W for 4 ohm. I am designing for a specific pair of speaker, the JMLab 913 that is 4 ohm. It is more important for me to get 8 to 10W of Class A than high power.
I am going to look at OS circuit. I just need to get to it. I just successfully run the FFT in LTSpice today.
Last edited:
Well, it's not just an opposite transistor. CFP is a completely different story (comparing to EF), as it involves the local feedback. It sounds good, but rather tricky in terms of local stability. It is practically paralleled with the number of pairs not more than 2 - either having 1 driver for both output transistors, or having 2 separate drivers.
I used twin-CFP arrangement in one of my designs, but it required careful simulations and additional tuning on the prototype. Read more material on the internet with regards to "CFP" or "Sziklai" to understand how it works and... handle with care 😉
I just corrected some mistake and simulated Cordell's circuit with 10K resistor. I did not use OPS, instead I just use the output of the VAS to drive the feedback resistors to get the GNFB. Output is 35Vpeak out of +/-40V rail. Looks pretty good to me.
Hi Alan
Best results (without resort to tedious manual matching) from this topology will require a quad matched NPN/PNP in a monolithic package. I'm not adverse to using surface mount parts provided the packages aren't unreasonably small.
Perhaps you would like to investigate whether such a package exists, preferably with high gain, low noise transistor elements ideally suited to this application. If we can find a suitable part then I would seriously consider respinning the PCB suit.
Best results (without resort to tedious manual matching) from this topology will require a quad matched NPN/PNP in a monolithic package. I'm not adverse to using surface mount parts provided the packages aren't unreasonably small.
Perhaps you would like to investigate whether such a package exists, preferably with high gain, low noise transistor elements ideally suited to this application. If we can find a suitable part then I would seriously consider respinning the PCB suit.
There is not quad that match that well I know of. In fact, the quad 2222 etc only said it's 4 individual transistor, nothing mention about on the same die and have any sort of match. I looked around this is what I have for the price and match on dual:
SSM2212RZ-R7 Analog Devices Inc | SSM2212RZ-R7CT-ND | DigiKey
SSM2220PZ Analog Devices Inc | SSM2220PZ-ND | DigiKey
The issue is the beta of the PNP is only 80min and the NPN is like 300. But thay match.
But, BUT!!!! I really think you can use a trim pot and trim everything together. You don't even need to waste your boards. Put the trim pot in place of one of the emitter resistor of the current mirror and see. You have the parts, it's easy to try. Trim it good, then use a hair dryer to heat up the board a little and see how the current drift!!!
Last edited:
Yeah I actually had in mind the THAT 340, but the Analogue Devices' parts look good too.
http://www.mouser.com/ds/2/407/300data-180007.pdf
CBS, do you have made any designs with balanced LTP input stages with current mirrors. in place of the more common resistive loads? I can't remember what you used as the front-end for your Fairchild VFET with EC design.
Alan, you sound pretty keen; if you'd like you can send me your address and I'll mail the boards to you.
http://www.mouser.com/ds/2/407/300data-180007.pdf
CBS, do you have made any designs with balanced LTP input stages with current mirrors. in place of the more common resistive loads? I can't remember what you used as the front-end for your Fairchild VFET with EC design.
Alan, you sound pretty keen; if you'd like you can send me your address and I'll mail the boards to you.
Thanks, I am very grateful. I sent you a PM already. I just did some more simulations. I am absolutely surprised the result of using a simple 2K resistor load looks that good too!!! Here are both LTSpice files for you to play with and the result.
"IPS VAS 1" is simulation of Cordell's circuit with 10K resistor.
"Normal IPS VAS with 2K resistor load" is simulation with just a simple 2K resistor load.
I also attached result of both simulations at 20KHz, run time set at 2.5mS, sampling interval at 0.009uS. So my result should be consistent for both. You can see the one with 2K resistor is just as good!!!! It would be very easy for you to jumper your board to just have the 2K resistor. You better double check my simulation though as today is the first day I learn how to do FFT!!! I yet to do simulation at 1KHz and low frequency. But 20KHz is supposed to be the worst case.
Thanks
Alan
That's a good one. Too bad the PNP has very low beta, or else I can set up the cascode on both sides at 15V to keep the total voltage at 30V and use one of the 2NPN/2PNP chip to do the whole complementary stage.
I used them in THIS experimental amp. As I recall, it worked great! 😎 The THAT matched arrays do not have high DC Hfe, I measure <90 for both polarities. No need for degeneration resistors though, high Ft and low noise.🙂 I have not used the Analog Devices components, they might be cheaper. THAT arrays are not so cheap.😱 I have also played with some matched mosfet arrays from Advanced Linear Devices called E-pad mosfet arrays. They can be programed to have certain specific Gm and other characteristics sort of like a 'linear EE-prom'🙂, although the ones I used were pre-programed, designated as ALD1105. Because of the low breakdown voltage, I cascode them with highish Goss J-fets.😉 I do not have the programming apparatus required but I think it would be real interesting to be able to use these in some sort of custom job.
Last edited:
Thanks Alan, PM received.
Actually I am not at all surprised that the simple resistor load performs extremely well. At this level we are well and truly gilding the lily, adding enormous complexity for marginal reductions in THD, that are already well below the threshold of human hearing. For an instrumentation amplifier in a piece of test equipment, where precision is required, then absolutely. For an audio amplifier its a moot point, specially when these "perfect" circuits will probably have a 10-100X more distortion in the real world due to layout, grounding, power supply and RFI ingress issues.
You seem to like fully balanced topologies; might I suggest you take a look at Bonsai's fully documented e-amp project:
Ovation e-Amp: A 180 Watt Class AB VFA Featuring Ultra Low Distortion
Best of all this is actually a built and tested project, not just a theoretical sketch in a book that may or may not work.
Actually this is only one of three IPS VAS I am going to do. I modularize the design that I can swap different IPS VAS in. The other one I like is the Apexaudio design. I am also thinking about a basic Blameless or Threshold design. If Nelson Pass made that famous, there got to be a reason.
Good plan. If you want a very low-THD design, and aren't obsessive, then the basic Blameless with EF3 should give you all the performance you need. The issues there are rail sticking and overload recovery, and there are well documented ways to minimize these problems for very little added complexity. People say the asymmetric VAS clipping can produce DC at the output; a well designed DC detection and solid-state disconnect circuit is the line of defence, and should be stock standard for any DC coupled amplifier in any case, in my opinion.
- Status
- Not open for further replies.