New Amplifier Project

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Hello everyone,

My current project is building a pair of amplifiers in a couple of empty Bryston 4B cases. These cases, along with a 3B case, were originally purchased from Bryston as display props for a dealer, and from the outside look like a complete working amplifier, complete with heatsinks, etc.

I have Acoustat model 2 electrostatic speakers, and have always been impressed by the way that PS Audio amplifiers have seemed to work well driving these speakers. For many years I have used a pair of PS model 2 power amplifiers which I heavily modified with new power supplies and upgraded parts, and have been quite satisfied with them, other than needing a bit more power. So I decided to build a better version in the 4B cases.

The circuit that I have been planning to use is essentially the same as a PS Audio model 2CPLUS amplifier, the last type in that series. I will be using MJ15024/25 for outputs, and will be paralleling 4 of each per side. Each 4B amplifier will be about 100 watts/8 ohms stereo, bridgeable to 400W mono. The amplifiers will also use a completely separate power supply for the driver section, as PS audio originally did. Resista resistors, wondercaps, and 4% silver solder will also be used throughout the amps. I have a prototype channel up and running, and have a few questions:


What is the function of the diode and resistor feedback network to the 2N3405 bias transistor? Is this good thing?

Can the gain of the amplifier be increased slightly? The prototype required well over 1 volt rms input for full output.

I've noticed that the feedback capacitors on amplifiers tend to be low voltage electrolytics, is there some advantage to using a 6.3V 470uF here as opposed to a 35V 470uF?

Any recommendations for driver transistors? I am currently using TIP41/42.

When the zobel network is connected, the amplifier will oscillate at a very high frequency. Without it it works fine, even when driving the electrostatics. Any suggestions as to what to look for here?

I feel that a separate regulated power supply for the driver circuitry is very important and results in a more stable and better sounding amplifier. This may be why this type amplifier works well with the difficult impedances and current demands of the electrostatic speakers. How do others feel about this?

I currently do not have a spice program, so if somebody would like to check the design that way I would certainly be interested in the results.

Any other recommendations for improving this circuit without getting too radical?

All suggestions and input are very much appreciated!

Thanks,
Doug:) :)
 

Attachments

  • psamp.gif
    psamp.gif
    9.3 KB · Views: 2,985
Hi,

I'm not an expert or anything coming near it so, don't take my word for granted and someone please correct me if i'm wrong.
:rolleyes:

Can the gain of the amplifier be increased slightly? The prototype required well over 1 volt rms input for full output.

With a current mirror on the differential pair you will increase de open loop gain of the amplifier, and also increase the voltage swing.

I've noticed that the feedback capacitors on amplifiers tend to be low voltage electrolytics, is there some advantage to using a 6.3V 470uF here as opposed to a 35V 470uF?

Not that i know of except the 6.3V cap being cheaper. I'd stick with the 35V one.

When the zobel network is connected, the amplifier will oscillate at a very high frequency. Without it it works fine, even when driving the electrostatics. Any suggestions as to what to look for here?

You probably have an oscillation somewhere, even without the zobel. Use a scope and you'll probably see it clearly even with a purely resistive load. Though it may work without the zobel I'd try and correct the problem first.
By the way, what is the speaker's impedance you're using? if it's 8ohm try replacing the zobel components with 100pf and 10ohm.

Regards,

Tiago
 
If you really have no oscillation without the zobel you can probably live without it.
Also, please try connecting with a R of 10ohm and C of 100pf, R connected to output and C to ground(I am still assuming a load of 8ohm). At what frequency does it oscillate? How is your grounding set up?
 
Gain stuff :)

Hello Doug,

Snip<
With a current mirror on the differential pair you will increase de open loop gain of the amplifier, and also increase the voltage swing.
>Snip

Yes you will increase the open loop gain (and maybe make a nice oscillator), however it will *NOT* change the closed loop gain of the amp.

To change the gain of the amp, just think of it like it's an opamp in a non inverting configuration, and change the feedback circuit to get the gain you need.

\Jens
 
Hello Jens,

the idea was having more open loop gain and then change the feedback network to have the same amount(or a little more) of feedback and still have more closed loop gain.
Increasing gain by reducing the feedback doesn't seem a very good idea.

Then again, i might just be wrong :smash:

Regards,

Tiago
 
Hi Doug

Basing on the schematics, one of the most obvious feature is an amplifier with quite a restricted bandwidth.

330pF + 23.1K at the input diff already sets the upper limit at 20KHz. Coupled with 750pF at the drivers, I would not be surprised if it measures -3dB at 20KHz.

The other prominent point is the larger than normal amount of negative feedback used. The voltage gain is set at 21.5x which works out to about 27dB. That accounts for "The prototype required well over 1 volt rms input for full output."

Going into the details, it would be much easier to set the diff current with a fixed resistor at the current source (sink actually) rather than to try to adjust it with a 5K preset.
 
The function of the diode and the resistor in the bias part seems to be there to increase the bias current for positive or negative going signals. Sliding bias. If the output is at ground potential the 1M resistor does nothing. For positive outputs the base of the bias transistor goes up and the 1M and 221k resistors are pulling it down a bit, trying to decrease Vbe and as a consequence of feedback by the 6k81 resistor the Vce will increase increasing the output bias current. A similar thing happens if the output goes down. Then the 221k resistor would pull up the base of the bias transistor via 1M, but that is prevented by the diode and the 91k4 resistor. These are pulling down and create a voltage divider with the 221k resistor to drive the base via 1M. Again bias current increases.

To increase the closed loop gain a little, just make the 2k21 resistor of the feedback network a little smaller. You have open loopgain enough.

No real advantage in using a 35V cap in the feedback network, 6.3V will do. If you want to make it better: put two caps of 1000u in series with minus to minus and use the plusses as the actual connections. I.e. make it bipolar.

Is your Zobel going to the star earth point? The oscillations suggest it is lifting the ground somewhere.


Steven
 
mcp said:
Hi Doug

Basing on the schematics, one of the most obvious feature is an amplifier with quite a restricted bandwidth.

330pF + 23.1K at the input diff already sets the upper limit at 20KHz. Coupled with 750pF at the drivers, I would not be surprised if it measures -3dB at 20KHz.

The other prominent point is the larger than normal amount of negative feedback used. The voltage gain is set at 21.5x which works out to about 27dB. That accounts for "The prototype required well over 1 volt rms input for full output."

Going into the details, it would be much easier to set the diff current with a fixed resistor at the current source (sink actually) rather than to try to adjust it with a 5K preset.

The 330p + 23k1 do not create a pole at 20kHz. It's just a gentle lead lag network in combination with the 3k01 collector resistors. Also the 750p caps are not decreasing the bandwidth to a sub 20kHz. Remember, they mainly affect the open loop gain. The closed loop gain can still be far above 20kHz, depending on the gain margin.

Just be careful in adjusting the 5k preset. Turning to zero may blow some transistors. It is used to adjust the tail current of the diff stage in order to make the output offset voltage close to zero. The tail current should be such that the voltage across both the 3k01 collector resistors is around 4 diode drops, like in the BD912 emitter.

Steven
 
What is the objective of the parralleled caps at the input, 8uF + .001uF + .68uF ? I understand the 8uF all by itself but the others? Was the orginal anticipating the 8uF would be some kind of electrolytic that didn't have flat bandwidth all the way out to 20kHz? Then there is the 150pF cap that filters HF. So first you let the HF pass then filter it off? If my WAG that it is a babdwidth issue reflecting limitations of ancient component why not just replace all three with one good quality 10uF metal film cap?
 
As long as I'm asking questions: What is the intended benefit of a sliding bias? I tried stiking the diode and resistors in to an existing SPICE model to see what the effect was. Vusually, any before/after traces I plotted on any scale just sat on top of each other. The "after" subtracted .00006 from the THD figure. Not earth shattering if true and just as likely to be an artifact of the order in which the emgine does it's calculation. Or simple attrition of significant digits.

Assuming N.P. wouldn't put it in a design just for fun, is it a way of making optimum bias easier to set and to maintain under real conditions?
 
Thank you to everyone who has posted responses so far, they have been very helpful!

When I removed my ammeter from the negative rail and cleaned up the wiring a bit the oscillations stopped! Doh! :xeye:

To increase the closed loop gain a little, just make the 2k21 resistor of the feedback network a little smaller. You have open loopgain enough.

OK, I reduced it to 1R5, and the gain is fine now. But, will this effect the gain when I use the inverting input on one channel when bridging?

What is the objective of the parralleled caps at the input, 8uF + .001uF + .68uF ? I understand the 8uF all by itself but the others? Was the orginal anticipating the 8uF would be some kind of electrolytic that didn't have flat bandwidth all the way out to 20kHz? Then there is the 150pF cap that filters HF. So first you let the HF pass then filter it off? If my WAG that it is a babdwidth issue reflecting limitations of ancient component why not just replace all three with one good quality 10uF metal film cap?

The original design used a 0.1uF film cap in parallel with a 10uF electrolytic. I am using original vintage IAR Wondercaps for these amplifiers and that is their recommended configuration for signal bypassing. I have always felt that multiple bypassing like this is best.

As long as I'm asking questions: What is the intended benefit of a sliding bias?

Yes, I'd like to know the answer to this as well. Should I take out the circuit or leave it in? :confused: :confused: My older PS Audio amplifiers do not have it.

Does anyone have any comments on the separate driver/output power supply topology?

So far it looks like I'm getting 125 watts into 8 ohms, with a frequency response extending reasonably flat to at least 50kHz. :cool:

Thanks again, everyone!!!
 
Originally posted by Steven

The function of the diode and the resistor in the bias part seems to be there to increase the bias current for positive or negative going signals. Sliding bias. If the output is at ground potential the 1M resistor does nothing. For positive outputs the base of the bias transistor goes up and the 1M and 221k resistors are pulling it down a bit, trying to decrease Vbe and as a consequence of feedback by the 6k81 resistor the Vce will increase increasing the output bias current. A similar thing happens if the output goes down. Then the 221k resistor would pull up the base of the bias transistor via 1M, but that is prevented by the diode and the 91k4 resistor. These are pulling down and create a voltage divider with the 221k resistor to drive the base via 1M. Again bias current increases.

Hello Steven

It did not occur to me at first that it is a Sliding Bias. My initial impression was that it shuts down the drivers in the event the outputs went full positive or negative rails.

You are correct about the 330p + 23k1. I have given the impression that it sets the dominent pole. It is indeed a phase lag compensation network set at 20KHz. Thank you for highlighting this.

As for the 750pF at the drivers, the slew rate of the amplifier will be quite severely affected, leading to my conclusion of a bandwidth limited, high negative feedback design.
 
The Peasant said:
..... What is the function of the diode and resistor feedback network to the 2N3405 bias transistor? Is this good thing?
As shown, I don't understand the function. The diode appears to be backwards, if this is an auto bias circuit, ala Nelson Pass A40.

The Peasant said:
..... Can gain be increased? The prototype required well over 1 volt rms input for full output.
That is normal and customary. Most preamps will drive it to full output easily. The gain is set at about 22 by the ratio of the 47K5 and 2K21 resistors, but you can change the ratio to say 30 by changing the 2K21 resistor to 1K6 (if you don't mind additional distortion). Personally, I would leave it alone, it's fine as is.

The Peasant said:
..... I've noticed that the feedback capacitors on amplifiers tend to be low voltage electrolytics, is there some advantage to using a 6.3V 470uF here as opposed to a 35V 470uF?
Even a 3v rating is fine and keeps physical size down. The voltage here should be almost zero in normal operation. However, the cap polarity is backward for the NPN input pair shown, as this cap will always have a slight negative voltage on it (maybe -300mv).

The Peasant said:
..... Any recommendations for driver transistors? I am currently using TIP41/42.
MJE15030/MJE15031 is much better pair. You're on the hairy edge of needing a predriver stage, given 4 output pairs. By the way, you need at most 2 pair for 100 watts.

The Peasant said:
..... When the zobel network is connected, the amplifier will oscillate at a very high frequency. Without it it works fine, even when driving the electrostatics. Any suggestions as to what to look for here?

Probably your Zobel network is putting high frequency into your signal ground circuit. Put the Zobel off card, right on the output binding posts is ideal. Then the ground current returns back to the star ground, not modulating your input and feedback signals. (You do have a separate return path for them, right?) Try that first and see if it helps.

There might be stability problems also if it oscillates after connecting the Zobel net. Hmm.... You might try an 8-10 ohm resistor in that net, but if moving the Zobel net off card does not help, I suspect you need to add a compensation capacitor on the VAS in addition to moving the Zobel net off card.

The usual practice to ensure stability would be to place the main compensation cap from base to collector of the VAS, here the BD912 on the diode string. The typical value is 50-100pF.

The Peasant said:
..... I feel that a separate regulated power supply for the driver circuitry is very important and results in a more stable and better sounding amplifier. This may be why this type amplifier works well with the difficult impedances and current demands of the electrostatic speakers. How do others feel about this?
My general philosophy is to try to make the front end immune to power supply noise, and use an unregulated supply with a modest RC filter for the front end (10 ohm, 100mf paralleled with your favorite film cap). The exception would be in driving MOSFETs, where there is significant voltage loss due to bias and gate drive requirements. Then a supply about 10V higher is advantageous.

But in the EF BJT case here it is risky. As has been pointed out elsewhere, if you aren't careful you will overdrive and saturate the outputs on clipping, and they will cross-conduct and self destruct. Baker clamps will prevent this.
 
Doug,

maybe you should move away those 750 pF and see if you still have oscillation.
Those capacitors may give a too large phase lag before you have unity gain which is the same as oscillation.

If you still experience oscillation:
Afterwards you can, as pointed out by others, put a Miller cap between B & C ob the BD912 VAS stage, but start with a much smaller value than 750 pF!

Tell us how it developed if/when you have tried it.


SAM9, (Doug),

I think you know that every capacitor have it's on fo, the smallest one on the input, 1 nF, do have propably a lower fo than 150 pF.
But on the other hand, I think 1 nF in the input cap is overkill, by putting a good 100 nF the Fo is still very high, probably couple of MHz.
This would be a better choice if RF is a concerne and the 150 pF will see a higher input impedance and make the work better if the smallest input cap is larger, eg. 100 nF.


Regards ;)
 
Oh yes, a 1uF cap across the 103R/2W resistor between the BD911 and BD912 in the output stage will probably reduce crossover distortion. The way I understand it is that when one driver turns "on" as the signal starts to swing in it's direction, it forces the other driver "off" by "stealing" current from it, thus making the transition between upper and lower halves of the output section cleaner.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.