I've got a dead Sansui 5000A with the incendiary F-1040 boards. One of them looks to have burnt, been repaired, and burnt again.
What can we build on the same footprint PCB as a drop-in replacement (or close) to bring this old amp back to better-than-new?
Here's my idea: take a standard Wolverine clone. Omit the SOA protection; there's not enough real estate on the boards. Include a simple time-based SSLR that unmutes a few seconds after power on and mutes instantly at power-off. Otherwise I'll scare my dog, he doesn't like thumping amps.
The most risky/innovative piece is the SSLR. I'm trying to do this on the cheap without an opto-isolator. The theory of operation of the SSLR is:
- R14, R15, R16 and zener D3 form a biasing network; at DC Vgs is set by the zener at 12V.
- Cap C3 keeps Vgs stable even as the amp plays, with both ends of this cap following the signal.
- R18 gracefully precharges and discharges the output cap.
- Qp1 drops Vgs instantly at loss of AC and discharges C3 in ~100ms (through R14) so that if power is lost and quickly restored, the relay must recharge for a bit before closing again.
- R20 / D4 assist with discharging C3; otherwise Qp1 would simply pull the negative end of C3 below ground without actually discharging it.
- R16 is there to pull the mosfet source nodes down when first setting up a positive Vgs at power on.
- D2 and C7 ensure that Qp1 can keep pulling the relay open until a few seconds after B+ has fallen to ground.
What can we build on the same footprint PCB as a drop-in replacement (or close) to bring this old amp back to better-than-new?
Here's my idea: take a standard Wolverine clone. Omit the SOA protection; there's not enough real estate on the boards. Include a simple time-based SSLR that unmutes a few seconds after power on and mutes instantly at power-off. Otherwise I'll scare my dog, he doesn't like thumping amps.
The most risky/innovative piece is the SSLR. I'm trying to do this on the cheap without an opto-isolator. The theory of operation of the SSLR is:
- R14, R15, R16 and zener D3 form a biasing network; at DC Vgs is set by the zener at 12V.
- Cap C3 keeps Vgs stable even as the amp plays, with both ends of this cap following the signal.
- R18 gracefully precharges and discharges the output cap.
- Qp1 drops Vgs instantly at loss of AC and discharges C3 in ~100ms (through R14) so that if power is lost and quickly restored, the relay must recharge for a bit before closing again.
- R20 / D4 assist with discharging C3; otherwise Qp1 would simply pull the negative end of C3 below ground without actually discharging it.
- R16 is there to pull the mosfet source nodes down when first setting up a positive Vgs at power on.
- D2 and C7 ensure that Qp1 can keep pulling the relay open until a few seconds after B+ has fallen to ground.
Attachments
Could we use the Sansui's stock transformer -- with no center tap on the secondary -- and drive a dual-rail DC-coupled amplifier from it? This sounds idiotic but bear with me.
In any normal direct-coupled class AB amp the center tap is grounded, this causes B+ and B- to be equal and opposite potential from ground.
With no center tap, the Sansui's power supply can only push B+ and B- to be 80V apart. The stock design ties B- to ground so B+ sits around +80V -- a single supply rail.
If we untie B- from ground, could we get balanced B+ and B- from this transformer? This is insane, but you could do it with a servo that keeps the amp's DC offset at the midpoint between the supply rails. If the rails drift too high, the DC offset goes positive, and the bias across the load pulls the rails back down (and vice versa.)
This is insane because the power supply isn't self-contained. It relies on the amp, and the load. This must work for a wide range of load impedances. High impedance loads are the most challenging, because the servo can only adjust the rails very slowly. It does settle out, eventually.
This simulates perfectly well! I call it the "better idiot". A Sansui F-1040 transmuted into 2015 ought to include some really harebrained idea that will probably blow up. Just like the original.
An advantage of converting to DC-coupled is that we only need two big filter caps, two fewer than the original. That opens up space in the chassis to fit another PCB. That could do protection and muting, it could also host the servos.
A disadvantage of this design is the filter caps should be 80V rated even though we normally expect them to carry only 40V. Likewise transistors and small caps used in this design could see the rails up to +/-80V and they should tolerate it.
The sim below shows the servo bringing the rails into DC balance in about 30 seconds, starting cold (skipping initial solution so all caps are initially discharged) with a 4 ohm load.
In any normal direct-coupled class AB amp the center tap is grounded, this causes B+ and B- to be equal and opposite potential from ground.
With no center tap, the Sansui's power supply can only push B+ and B- to be 80V apart. The stock design ties B- to ground so B+ sits around +80V -- a single supply rail.
If we untie B- from ground, could we get balanced B+ and B- from this transformer? This is insane, but you could do it with a servo that keeps the amp's DC offset at the midpoint between the supply rails. If the rails drift too high, the DC offset goes positive, and the bias across the load pulls the rails back down (and vice versa.)
This is insane because the power supply isn't self-contained. It relies on the amp, and the load. This must work for a wide range of load impedances. High impedance loads are the most challenging, because the servo can only adjust the rails very slowly. It does settle out, eventually.
This simulates perfectly well! I call it the "better idiot". A Sansui F-1040 transmuted into 2015 ought to include some really harebrained idea that will probably blow up. Just like the original.
An advantage of converting to DC-coupled is that we only need two big filter caps, two fewer than the original. That opens up space in the chassis to fit another PCB. That could do protection and muting, it could also host the servos.
A disadvantage of this design is the filter caps should be 80V rated even though we normally expect them to carry only 40V. Likewise transistors and small caps used in this design could see the rails up to +/-80V and they should tolerate it.
The sim below shows the servo bringing the rails into DC balance in about 30 seconds, starting cold (skipping initial solution so all caps are initially discharged) with a 4 ohm load.
Attachments
Last edited:
The thing is - it's not DC coupled. Rather it's capacitor coupled but to both rails.
The DC servo is a complete waste of resource and it can possibly result in LF instability - what you really need is an input high pass filter. The point of it is to prevent frequencies lower than HPF created by the output coupling to enter the amp as it could then 'float' it's own power rails too close to the ground reference, possibly resulting in all kinds of nasty behavior such as latch-up (which you should investigate in the sim). The turnover frequency of the input HPF must be higher than the output HPF, and quite a bit too.
There is a pertinent question here, and it's about what you hope to gain by this - the low end cut is still defined by an output cap, except now there are two, MAYBE half in size if the main filter cap is MUCH bigger than they are. Not only is the ouptut current going through the caps, but now so is the rectification current from the CPU - something which is ideally kept apart. Seems like more work and parts for same or worse result. At least with a single output cap, it is possible to include it at least partially into the NFB loop and mitigate some of it's effects on the output signal.
The DC servo is a complete waste of resource and it can possibly result in LF instability - what you really need is an input high pass filter. The point of it is to prevent frequencies lower than HPF created by the output coupling to enter the amp as it could then 'float' it's own power rails too close to the ground reference, possibly resulting in all kinds of nasty behavior such as latch-up (which you should investigate in the sim). The turnover frequency of the input HPF must be higher than the output HPF, and quite a bit too.
There is a pertinent question here, and it's about what you hope to gain by this - the low end cut is still defined by an output cap, except now there are two, MAYBE half in size if the main filter cap is MUCH bigger than they are. Not only is the ouptut current going through the caps, but now so is the rectification current from the CPU - something which is ideally kept apart. Seems like more work and parts for same or worse result. At least with a single output cap, it is possible to include it at least partially into the NFB loop and mitigate some of it's effects on the output signal.
ilimzn, thanks for your thoughts.
I believe the DC servo is critical. It keeps the rails at +40 and -40 as opposed to some other random values that happen to be 80V apart. Without the DC servo, even a miniscule DC offset at the output could send the rails far out of balance.
Granted there may be other ways of keeping the rails balanced. I suppose you could hang power resistors from B+ to gnd and gnd to B-. It would burn power and create extra supply ripple. Or maybe someone who knows how to build transformers could add the missing center tap. I popped the cover off the transformer to have a look, and the wires are buried beneath layers of glue and crumbly paper-mache looking stuff. It scared me right off of messing with it further.
You're right about LF instability being a concern. I saw LF oscillation in early versions of this circuit. The current topology isn't prone to it, you can make big changes to the values of C7, C8, R22, and R_load and not get self-sustaining oscillations. With certain values you can get peaks in the LF frequency response, and corresponding long settling times for the rails.
You were right about latch-up. If B+ fell to ground, the current source on the input stage wouldn't operate and the amp would be stuck there. I added Zener D1 to prevent B+ from falling below ground or exceeding 60V, and vice-versa for D2 and B-. That should be enough to prevent either rail from sticking to ground. On a variac at partial voltage, latchup is still possible so that could make bringup interesting.
There is an input high pass filter, it's R11 and C4.
Here's today's version. It adds 60V zeners D1 and D2, reduces the value of C8 so that the servo circuit starts working faster (in 1s versus 8s before), and increases C5 for better PSRR. It adds R14 so that the amp always sees some finite load and the servo can balance the rails (albeit slowly) when no speakers are connected. It also deletes the former C6 which wasn't helping PSRR, it was actually making it a bit worse.
I believe the DC servo is critical. It keeps the rails at +40 and -40 as opposed to some other random values that happen to be 80V apart. Without the DC servo, even a miniscule DC offset at the output could send the rails far out of balance.
Granted there may be other ways of keeping the rails balanced. I suppose you could hang power resistors from B+ to gnd and gnd to B-. It would burn power and create extra supply ripple. Or maybe someone who knows how to build transformers could add the missing center tap. I popped the cover off the transformer to have a look, and the wires are buried beneath layers of glue and crumbly paper-mache looking stuff. It scared me right off of messing with it further.
You're right about LF instability being a concern. I saw LF oscillation in early versions of this circuit. The current topology isn't prone to it, you can make big changes to the values of C7, C8, R22, and R_load and not get self-sustaining oscillations. With certain values you can get peaks in the LF frequency response, and corresponding long settling times for the rails.
You were right about latch-up. If B+ fell to ground, the current source on the input stage wouldn't operate and the amp would be stuck there. I added Zener D1 to prevent B+ from falling below ground or exceeding 60V, and vice-versa for D2 and B-. That should be enough to prevent either rail from sticking to ground. On a variac at partial voltage, latchup is still possible so that could make bringup interesting.
There is an input high pass filter, it's R11 and C4.
Here's today's version. It adds 60V zeners D1 and D2, reduces the value of C8 so that the servo circuit starts working faster (in 1s versus 8s before), and increases C5 for better PSRR. It adds R14 so that the amp always sees some finite load and the servo can balance the rails (albeit slowly) when no speakers are connected. It also deletes the former C6 which wasn't helping PSRR, it was actually making it a bit worse.
Attachments
- Status
- Not open for further replies.