Since we're discussing this circuit's operation, I do have a few questions:
- What is the purpose of the 22K R4 resistor that bypasses the emitter and collector of Q2? I haven't seen this used in other CCSs.
- Regarding slew rate, if I run a 0.5 amplitude 20Hz square wave, I get a slew rate of around ±15V/μs. If I remove the 150pF input cap C2, this increases to over ±40V/μs. This input filter seems to be more aggressive than I normally see and appears to be limiting slew rate. Is this intentional and specific towards the amps sonics? At the values of 4.7k & 150pF, the -3db response is 226kHz. Lower still if the source impedance is high. If I drop C2 to 47pF, the corner is 720kHz and the slew rate doubles to around ±30V/μs.
- The value for R3 at 4.7K seems to be use case specific. If you need to guard against unknown and potentially high source impedances, this makes sense. I only mention this because R3 at 4.7K is likely the dominate source of noise. Unless a high source impedance is anticipated, I would think dropping this to 1K would be reasonable. And C2 can then be adjusted accordingly.
Something like 3V/µs is plenty for music, you won't encounter anything more in real life. Douglas Self wrote about slew rate requirements in his excellent book. The input filter seems reasonable and pretty standard at around 200kHz and is there to keep any high frequency cruft and EMI stuff away from your speakers.
I'm with you there from the noise perspective, though. R3 is absolutely the main offender, and I have changed the input filter to 470R and 1nF for my planned build.
I'm with you there from the noise perspective, though. R3 is absolutely the main offender, and I have changed the input filter to 470R and 1nF for my planned build.
1/ I honestly can not remember what the reasoning was now. Twenty years down the line there are definitely things I would do very differently
2/ and 3/ Limiting the bandwidth at the input to the amp is sensible and the response is still very extended. My amp and preamp used a 10k motorised ALPS pot feeding the power amp.
Kudos to Mooly for an interesting design!
This is my first post in this thread and I'm always fascinated by bias design and DC servos. I may be able to suggest a remedy for the settling dynamics of the servo. The changes are modest and only require a couple of added resistors.
The heart of the servo is the integrator formed by U1, C3, and R27. I believe the ringing arises from how the integrator time constant interacts with amplifier low-frequency gain-roll-off arising from C6,R19 and C1,R21. So, to essentially eliminate variation in amplifier gain seen by the integrator, I propose the following modifications:
1. Q1 base bias is established by connecting R21 to ground rather than to the servo opamp output. Consequently, Q1 emitter will be about +0.6V
2. Establish servo adjustment of bias by connecting a new 22k resistor from the opamp output to Q1 emitter. (22k is equal to R18, yielding unity gain to the integrator.)
3. Install a new 82k resistor from the +45V rail to supply current to Q1, R18, and the new 22k servo resistor. This resistor should connect at the junction of C6 and R19 to suppress ripple on the +45 rail.
82k estimated as follows:
Referring to Mooly's schematic in post 1442, it appears that Q1 bias current is the current flowing through R18, i.e. -5.0307651V/22k or about 0.23mA. Revised voltage across R18 is about 0.6V, or about 0.027mA. Voltage across 22k servo resistor is about 6.2V, or about 0.28mA. Sum of these three currents is about 0.54mA. So 45V/0.54mA = ~83.5k. Choose 82k.
Servo time constant is about C3*R27, or about 0.1s, or about 1.6Hz. Bias at C1 and C6 have to charge, but I do not foresee ringing behavior. It may be acceptable to speed the settling time by changing C3*R27.
This is my first post in this thread and I'm always fascinated by bias design and DC servos. I may be able to suggest a remedy for the settling dynamics of the servo. The changes are modest and only require a couple of added resistors.
The heart of the servo is the integrator formed by U1, C3, and R27. I believe the ringing arises from how the integrator time constant interacts with amplifier low-frequency gain-roll-off arising from C6,R19 and C1,R21. So, to essentially eliminate variation in amplifier gain seen by the integrator, I propose the following modifications:
1. Q1 base bias is established by connecting R21 to ground rather than to the servo opamp output. Consequently, Q1 emitter will be about +0.6V
2. Establish servo adjustment of bias by connecting a new 22k resistor from the opamp output to Q1 emitter. (22k is equal to R18, yielding unity gain to the integrator.)
3. Install a new 82k resistor from the +45V rail to supply current to Q1, R18, and the new 22k servo resistor. This resistor should connect at the junction of C6 and R19 to suppress ripple on the +45 rail.
82k estimated as follows:
Referring to Mooly's schematic in post 1442, it appears that Q1 bias current is the current flowing through R18, i.e. -5.0307651V/22k or about 0.23mA. Revised voltage across R18 is about 0.6V, or about 0.027mA. Voltage across 22k servo resistor is about 6.2V, or about 0.28mA. Sum of these three currents is about 0.54mA. So 45V/0.54mA = ~83.5k. Choose 82k.
Servo time constant is about C3*R27, or about 0.1s, or about 1.6Hz. Bias at C1 and C6 have to charge, but I do not foresee ringing behavior. It may be acceptable to speed the settling time by changing C3*R27.
Not working for me neither. Even with the supply for the opamp set to +-45V, it will swing to one of the rails and stick there. Somewhere around 44k/45k instead of 82k seems to be an unstable turning point.
Another approach would be to make R2=1M like R15 and increase C8 to 470nF. Reasonably fast settling time and the feedback network stays untouched (22k, 470R, 470uF).
Another approach would be to make R2=1M like R15 and increase C8 to 470nF. Reasonably fast settling time and the feedback network stays untouched (22k, 470R, 470uF).
Hmmm.
Just doodling. With the servo snipped out and replaced by a voltage source and now using a 0.33uF input cap and a 150uF feedback return gives a 'vinyl friendly' basic response.
Adding the servo back and increasing the 0.1uF to 1uF for the servo cap gives this settling time. 15 seconds in and the offset is around 30mv and after 20 seconds is essentially zero:
Output transient at power on:
Offset from 10 seconds onward. So the 0 point at the start is actual T+10 seconds and the finish is at 30 seconds from power on:
The final response with servo in place. Source impedance now has no notable effect beyond altering the amplitude (as expected as the source impedance in in series with the input impedance of the amp):
Using a 10meg and retaining the original 0.1uF will give the same result.
Just doodling. With the servo snipped out and replaced by a voltage source and now using a 0.33uF input cap and a 150uF feedback return gives a 'vinyl friendly' basic response.
Adding the servo back and increasing the 0.1uF to 1uF for the servo cap gives this settling time. 15 seconds in and the offset is around 30mv and after 20 seconds is essentially zero:
Output transient at power on:
Offset from 10 seconds onward. So the 0 point at the start is actual T+10 seconds and the finish is at 30 seconds from power on:
The final response with servo in place. Source impedance now has no notable effect beyond altering the amplitude (as expected as the source impedance in in series with the input impedance of the amp):
Using a 10meg and retaining the original 0.1uF will give the same result.
Hi Mooly,
Your mods in 1485 are exactly as I suggested, and I've been a complete idiot.
My thinking was that your original servo pulled base (and emitter) of Q1 negative to establish bias, so I'll just force the correcting current into the emitter and maintain the same feedback sense. WRONG! Q1 is essentially the front end of a current feedback amp--- the base is + input and the emitter is -input. My servo made positive feedback instead of negative.
I'm hugely embarrassed and apologize for wasting everyone's time.
Your mods in 1485 are exactly as I suggested, and I've been a complete idiot.
My thinking was that your original servo pulled base (and emitter) of Q1 negative to establish bias, so I'll just force the correcting current into the emitter and maintain the same feedback sense. WRONG! Q1 is essentially the front end of a current feedback amp--- the base is + input and the emitter is -input. My servo made positive feedback instead of negative.
I'm hugely embarrassed and apologize for wasting everyone's time.
You might like this one for looking at squarewave behaviour. The squarewave starts at zero volts and so the output is correct and swings equally in each direction. Alter the lines shown to set frequency and amplitude. TR is the risetime. The sim run time will alter automatically to suit whatever frequency you enter. The .tran line tells it when to start and stop saving data. For low frequency testing you can alter that to for example {1000/f} {995/f} to allow the circuit to settle.
This also works on the sim with the 'real' PSU but there you need to alter the above to more like 20000/F 19995/f to allow time for the rails to build.
10kHz
And at 20Hz showing the bass boost effect of that peak in the response.
And with the modified values for input cap, feedback cap and servo at 20Hz. That is exactly what you would see with a bass roll off:
So lots to play with for those interested in the sims
The squarewave generator file is also attached separately (and I can't claim credit for that one) can be copied from one LT sim to another as desired. Just open both in the same session, use the copy function to draw around it and then drag it to the toolbar at the top and open the sim it is to be placed into.
This also works on the sim with the 'real' PSU but there you need to alter the above to more like 20000/F 19995/f to allow time for the rails to build.
10kHz
And at 20Hz showing the bass boost effect of that peak in the response.
And with the modified values for input cap, feedback cap and servo at 20Hz. That is exactly what you would see with a bass roll off:
So lots to play with for those interested in the sims
The squarewave generator file is also attached separately (and I can't claim credit for that one) can be copied from one LT sim to another as desired. Just open both in the same session, use the copy function to draw around it and then drag it to the toolbar at the top and open the sim it is to be placed into.
You picked a rather high -3dB frequency of around 12Hz. A more common approach would be to go one magnitude above and below the audio band, just like you did with the input filter (20kHz -> 200kHz). That would be around 2Hz for the lower end. To keep the servo from interfering with the frequency response, its response will have to go down another order of magnitude, like 0.2Hz. This approach will lead to a pretty long settling time, albeit still being 'vinyl friendly'. Apart from that, you probably won't find anything close to 20Hz square waves in actual music.
R4 was in parallel with C4 and this would still carry a charge at switch off since the discharge route to earth is via 100k making it a case of how long does it take for the supply capacitors to discharge to a point where Q2 and Q3 vbe levels are below their conduction thresholds. This appears to be a circuit to mute spurious noises at power off. This involves more than one variable.
There are ways to disconnect a relay in the output line by sensing whether the secondary transformer winding is conducting or not. One can tell at an instant as there is an audible click when relays switch.
The CCS circuit in this has the same structure as in a design by Linsley-Hood published in Hi-Fi News and Record Review in the January issue of 1980. This is a single supply rail affair with an output capacitor with a value of 2200 uF and this needs to be charged slowly at turn on to avoid a large current draw in a short time. C4 will slow the switch on process for the The CCS due to the high value resistor at the bottom end to earth we don't need this so C4 should go. The other capacitor requiring to be charged at power up is C3. These capacitor designations are as per the original circuit in the image shown in post 1. This predecessor CCS to the current evolution should be seen in a different context to a direct coupled application.
I may be missing the point of the discussion but the servo circuit is quite able to work in 10kHz square wave simulations with C3 replaced a link to earth so it works at dc. There is an initial spike in the voltage at 0 seconds but it does not linger to any significant extent. The relay used to mute at turn on should also detect excess dc in the event of component failure.
Linsley-Hood published a full dc output circuit resembling that in Hi-Fi News which is quite remarkable. This was published in Electronics Today International in June 1975 in an article titled Amplifier Technology which may be of interest. If so I can supply the details.
Last edited:
They are interesting variations on the original. Removing the cap in the feedback return sort of feels wrong to me and not just because the gain within the amp is now maintained to DC
One variation I would look at again would be to dispense the current source altogether and go back to a traditional bootstrapped resistor arrangement.
One variation I would look at again would be to dispense the current source altogether and go back to a traditional bootstrapped resistor arrangement.
C4 will also help increase the PSRR a little, and "the high value resistor" R6 is needed for the CCS to work at all. I would not remove C4, and it does not have any noticeable influence on the turn-on settling time in this amp. Instead, I'd rather add an additional JFET current source in series with R6, as that will keep the output stage bias more constant with regard to rail voltage fluctuations or droop.
For the bootstrap, it would be interesting to try dadod's Bootstrap CCS that uses a depletion mode MOSFET for the lower resistor. In theory you get the sonics of a bootstrap without the current sensitivity to power supply variations. I've played with this in simulations and it seems to work well.
I think there is a case to take the servo down to dc although I have not ventured in that direction. Electrolytic capacitors are seen as OK provided these are sufficiently large in the level of capacitance and this can reduce THD percentages.
There was a move toward using plastic capacitors, which while bulkier can be small enough if circuit impedance in the nfb chain are increased by using high value resistors. JLH went down this road in 1984 having published a three part series of articles entitled "Audio Amplifier Design, Engineering or Alchemy" in Everyday with Practical Electronics from August to October 1983.
With respect to the dc blocking capacitor in the feedback arm to earth he noted that parasitic resistances and impedance in electrolytic capacitors will vary with shape of applied voltage, polarity frequency, and temperature. This combination of factors has little effect on THD. However when the amplifier is measured for IMD using a spectrum analyser driven by several simultaneous signals to track down the spurious signals having been generated, then imperfection differences in the type of blocking capacitor are more easily seen. It would be a good thing if SPICE models were available for simulations but that is a pipe dream.
Those who have built this project are in a position to run listening trials with or without the short circuit of the dc blocking capacitor. A shorting switch could be soldered to make short sequences of music either way.
One point to make about the output stage is that it has considerable voltage gain and if one looks at the currents driving this in simulations these are very small. That said the VAS stage is not subject to a significant load. Accordingly a buffer stage in front of this to bolster Q1 could be dispensed with.
I see there are no Zener diodes to protect the gates of the output MOSFET, pairs.
Last edited:
If anyone is wondering why I don't try this it is simply because the power amps in mine are not easily worked on, in fact 'the cans' which were originally from Sony SLC9 Beta VCR PSU's have not been opened since the amp was first constructed.
Simple answer to that was that I never envisaged any situation in domestic use that would bring them into play.