SebastianH
Member
Quick update: Both driver boards (Rev D) are ready for testing...
Initial results are fine.
More results soon.
Initial results are fine.
More results soon.
Sansui AU-717: Restoration with new boards
- Thread starter SebastianH
- Start date
-
- Tags
- pcb restoration sansui au 717
SebastianH
Member
A quick update with a couple measurements that you can explore yourself:
Measurements
Just an example:
What do you think? Anything you want me to measure (I can't promise to do it, but we'll see)?
I think its time to publish the gerber files for the driver boards. in a couple of days
Measurements
Just an example:
What do you think? Anything you want me to measure (I can't promise to do it, but we'll see)?
I think its time to publish the gerber files for the driver boards. in a couple of days
Great metrics on the Rev D boards! The Damping Factor around ~110 in the sub-1kHz region is outstanding and the L/R channel symmetry is spot on.
If you're taking requests for additional measurements, I’d love to see:
By the way, it’s fantastic news that you’ll be sharing the Gerber files in a couple of days! That is extremely generous of you and a huge contribution to the Sansui community. I’m definitely looking forward to downloading them and checking out your layout work.
Thanks again for sharing all these detailed test results!
If you're taking requests for additional measurements, I’d love to see:
- A 10kHz square wave response into a capacitive load (8Ω∥2μF) to check phase margin and stability.
- IMD (Intermodulation Distortion) figures (SMPTE or CCIF).
- What kind of Slew Rate (V/μs) are these new driver boards achieving compared to the stock design?
By the way, it’s fantastic news that you’ll be sharing the Gerber files in a couple of days! That is extremely generous of you and a huge contribution to the Sansui community. I’m definitely looking forward to downloading them and checking out your layout work.
Thanks again for sharing all these detailed test results!
SebastianH
Member
1. I did such tests at maybe 8Vpp or something (1W if it was a sine wave), but did not record any of it with the software. Pretty sure that I don't have 2uF, but likely 1uF/2.2uF film. It's certainly something I can do. I probably won't go too high with the amplitude, because I like my transistors and snubber resistors etc. But what did you have in mind?
2. Haven't implemented it yet. I don't think the QA40x API has an endpoint for this (but don't quote me), so I probably would have to load the raw fft data and calculate it. I do not have access to the actual standards, so I have to rely on what the LLMs tell me, implement and test, I guess.
3. Unfortunately I don't have the best documentation of the amp in its original form. My bench setup and software is getting better and better, so if I did this now, things would be a bit different... The slew rate of the new boards is about 32 to 36V/us. If I remember correctly the slew rate was similar.
Thanks! Letting the AI agents build the software was a ton of fun. And a bit scary as well. Letting claude control actual hardware (scope/function generator) is really, really weird. But I said: here's the scope's documentation. You can send scpi commands to the scope and find out exactly what the scope does. Now figure out how to implement the Time Series Map and THD Map measurements (same for the 34401A DMMs) and it did. Ok, a bit simplified and I had to help a bit every now and then (e. g. this memory depth is only available if setting x is set to y), but not that much. (Sonnet 4.6 at the time.)
A couple of screen shots (dark mode today):
Next weekend, I hope.
Thanks for the detailed feedback!
1. On the Square Wave / Capacitive Load Test: Regarding 'what I had in mind': I wasn't looking for high-amplitude stress testing—1W (or ~8Vpp) into an 8$,\Omega$ dummy load with a 1µF or 2.2µF film cap in parallel is more than enough to evaluate stability.
The main goal is just to check the rising edge on the scope for any excessive ringing, overshoot, or sustained parastic oscillation at the corners. A clean step response under a capacitive load confirms that the Phase Margin on your Rev D design is rock solid in real-world scenarios (where speaker cables and crossovers add reactive loads). No need to push high power at all!
2. On IMD: Understood on the QA40x software/API limitations. Don't sweat calculating it manually from raw FFTs unless you really enjoy the coding part—the THD and Damping Factor maps you already provided paint a fantastic picture of the board's performance.
3. On Slew Rate: 32 to 36 V/µs is a stellar figure! That proves the new driver topology retains all the fast, wide-bandwidth character that made the AU-717 famous, while offering a modern, stable layout.
That’s both mind-blowing and slightly terrifying! Having Claude auto-generate SCPI scripts to interface directly with your scope, function generator, and 34401A DMMs is next-level lab automation.
It’s wild how well these models understand instrument control syntax once you hand them the manual, even with those weird edge cases like memory depth dependencies. Talk about building your own custom Audio Precision setup on a budget!
The dark mode screenshots look super clean. Thanks for sharing a bit of the 'behind the scenes' on how you built this test suite!
The dashboard looks absolute professional-grade!
Seeing the Rise Time sitting between 1.52µs and 1.61µs at 75W into 8$\Omega$ is insane. That perfectly backs up the ~35V/µs Slew Rate you mentioned.
Also, having the software automatically shift test frequencies to avoid mains harmonics ('Avoid mains harmonics') is a brilliant touch. It saves so much manual headache when chasing real THD vs floor hum.
Thanks for sharing these screenshots, this is definitely one of the most thorough AU-717 testing setups I've ever seen on AudioKarma!
Looking forward to seeing the Gerber release whenever you get a chance to clean them up. Thanks again for doing such thorough engineering work on this!
1. On the Square Wave / Capacitive Load Test: Regarding 'what I had in mind': I wasn't looking for high-amplitude stress testing—1W (or ~8Vpp) into an 8$,\Omega$ dummy load with a 1µF or 2.2µF film cap in parallel is more than enough to evaluate stability.
The main goal is just to check the rising edge on the scope for any excessive ringing, overshoot, or sustained parastic oscillation at the corners. A clean step response under a capacitive load confirms that the Phase Margin on your Rev D design is rock solid in real-world scenarios (where speaker cables and crossovers add reactive loads). No need to push high power at all!
2. On IMD: Understood on the QA40x software/API limitations. Don't sweat calculating it manually from raw FFTs unless you really enjoy the coding part—the THD and Damping Factor maps you already provided paint a fantastic picture of the board's performance.
3. On Slew Rate: 32 to 36 V/µs is a stellar figure! That proves the new driver topology retains all the fast, wide-bandwidth character that made the AU-717 famous, while offering a modern, stable layout.
That’s both mind-blowing and slightly terrifying! Having Claude auto-generate SCPI scripts to interface directly with your scope, function generator, and 34401A DMMs is next-level lab automation.
It’s wild how well these models understand instrument control syntax once you hand them the manual, even with those weird edge cases like memory depth dependencies. Talk about building your own custom Audio Precision setup on a budget!
The dark mode screenshots look super clean. Thanks for sharing a bit of the 'behind the scenes' on how you built this test suite!
The dashboard looks absolute professional-grade!
Seeing the Rise Time sitting between 1.52µs and 1.61µs at 75W into 8$\Omega$ is insane. That perfectly backs up the ~35V/µs Slew Rate you mentioned.
Also, having the software automatically shift test frequencies to avoid mains harmonics ('Avoid mains harmonics') is a brilliant touch. It saves so much manual headache when chasing real THD vs floor hum.
Thanks for sharing these screenshots, this is definitely one of the most thorough AU-717 testing setups I've ever seen on AudioKarma!
Looking forward to seeing the Gerber release whenever you get a chance to clean them up. Thanks again for doing such thorough engineering work on this!
SebastianH
Member
Thank you! Regarding the step response with 8ohm || 2.2uF. I posted one scope screenshot from a previous board revision:
There was massive overshoot, and some ringing (1uF), but it was stable/did not oscillate. Back then I tested it with 2.2uF as well and it was substantially worse, i.e. more overshoot and more ringing, but it was still stable/did not break into oscillation. I would be surprised if any fast amplifier like this would have a truly clean step response like without a capacitor in parallel. But I guess you meant something like this?
The Rev D boards should behave very similar. I have to repeat the test with the right channel though to be sure.
I think claude will get a chance to implement the IMD measurement...
There was massive overshoot, and some ringing (1uF), but it was stable/did not oscillate. Back then I tested it with 2.2uF as well and it was substantially worse, i.e. more overshoot and more ringing, but it was still stable/did not break into oscillation. I would be surprised if any fast amplifier like this would have a truly clean step response like without a capacitor in parallel. But I guess you meant something like this?
The Rev D boards should behave very similar. I have to repeat the test with the right channel though to be sure.
I think claude will get a chance to implement the IMD measurement...
Thanks Sebastian! Yes, that scope screenshot from Feb 18th is exactly what I had in mind.
Staying completely stable without breaking into sustained oscillation under a 2.2µF reactive load is the ultimate test for a fast wide-band topology like the AU-717. As you mentioned, expecting zero overshoot/ringing under such heavy capacitive loading on a high-speed design would be unrealistic—the key takeaway is that the phase margin holds up and it stays rock-solid.
Awesome to hear that Claude might get to work on the IMD measurements too! Take your time with the Rev D tests and enjoy the process.
Staying completely stable without breaking into sustained oscillation under a 2.2µF reactive load is the ultimate test for a fast wide-band topology like the AU-717. As you mentioned, expecting zero overshoot/ringing under such heavy capacitive loading on a high-speed design would be unrealistic—the key takeaway is that the phase margin holds up and it stays rock-solid.
Awesome to hear that Claude might get to work on the IMD measurements too! Take your time with the Rev D tests and enjoy the process.
By the way, while catching up on the earlier pages of your build thread, I noticed your custom F-2663 Power Supply & Protector board, along with your custom MOSFET SSR design. That relay module is an absolute work of art!
Regarding the F-2663 board: if you ever decide to share the Gerber files for that main PSU/Protector PCB alongside the Rev D driver boards, I’d be extremely interested. No pressure at all, of course—just wanted to compliment the layout work on it as well!
Regarding the F-2663 board: if you ever decide to share the Gerber files for that main PSU/Protector PCB alongside the Rev D driver boards, I’d be extremely interested. No pressure at all, of course—just wanted to compliment the layout work on it as well!
SebastianH
Member
quick update:
I added a test with 8 ohm || 2.2uF. As expected the amp is stable. With a very high amount of confidence the overshoot and ringing comes mainly from the resonant circuit formed by the output inductance (output inductor + internal wiring, external speaker wire to the load resistor + other parasitic inductance) + the load capacitance (+...).
Quick LT spice sim:
- assumed 250n for internal wiring
- assumed 2u for the output inductance. If I remember correctly I measured a higher inductance than the specified 1.5uH for the original coil. But couldn't tell you at which frequency I measured it etc.
- assumed 500n for the speaker cable going to the load (~0,5m 0.75mm^2 speaker cable, not particularily great, but whatever )
Did the quick simulation of the amplifier as well:
At the emitter resistors vs. at the load.
Actual measurement at the load:
You can look at it yourself: Stability measurements
The simulation predicted a peak voltage of 8.39V, the measurement shows about 8.8V (left), 8.4V (right).
The simulation predicted a frequency of 63 kHz, the measurement shows about 64 kHz.
Why does the sim come so close? Simple answer: I started with what I know and tweaked the parameters a bit. It's not a proof of anything, really
But it's close enough to conclude that the amplifier works as expected, in my opinion.
IMD is another rabbit hole. Haven't yet verified the results (i. e. the actual values) too thoroughly, but I did a sweep (IMD measurements) from 0.1 W to 58W (the SMPTE signal has a crest factor of about 1.715, so the amp would clip hard at 85W). The spectrum shows... not much distortion...
Slewrate was already in the data set, so nothing new to report.
Not generally opposed to. But there is the issue with Sansui's strange multivibrator implementation that in my opinion isn't the greatest idea at all. If someone can shed some light on why the original circuit is supposed to work without killing the transistors then please do... If I remember correctly, someone replaced these exact transistors on my original board. Coincidence? Maybe. One thing is clear to me: My replacement transistors did not like to be in this circuit and I wouldn't want to be in there either
Have two options: reverse protection diode (my current bodge for Rev A, don't like it tbh) or completely avoid this circuit and use something fancy like the 555 timer for the blinking (was available at the time if that matters). (Btw. there is not to much space on the board to work with, so no complex discrete circuits...)
Haven't decided yet, I've done two variants (KiCAD only, not actually built).
btw.: These measurements took quite some time, so I still have to fix some CI/CD stuff for the gerber exports and prepare some documentation for the driver boards...
I added a test with 8 ohm || 2.2uF. As expected the amp is stable. With a very high amount of confidence the overshoot and ringing comes mainly from the resonant circuit formed by the output inductance (output inductor + internal wiring, external speaker wire to the load resistor + other parasitic inductance) + the load capacitance (+...).
Quick LT spice sim:
- assumed 250n for internal wiring
- assumed 2u for the output inductance. If I remember correctly I measured a higher inductance than the specified 1.5uH for the original coil. But couldn't tell you at which frequency I measured it etc.
- assumed 500n for the speaker cable going to the load (~0,5m 0.75mm^2 speaker cable, not particularily great, but whatever
)Did the quick simulation of the amplifier as well:
At the emitter resistors vs. at the load.
Actual measurement at the load:
You can look at it yourself: Stability measurements
The simulation predicted a peak voltage of 8.39V, the measurement shows about 8.8V (left), 8.4V (right).
The simulation predicted a frequency of 63 kHz, the measurement shows about 64 kHz.
Why does the sim come so close? Simple answer: I started with what I know and tweaked the parameters a bit. It's not a proof of anything, really
But it's close enough to conclude that the amplifier works as expected, in my opinion.IMD is another rabbit hole. Haven't yet verified the results (i. e. the actual values) too thoroughly, but I did a sweep (IMD measurements) from 0.1 W to 58W (the SMPTE signal has a crest factor of about 1.715, so the amp would clip hard at 85W). The spectrum shows... not much distortion...
Slewrate was already in the data set, so nothing new to report.
Not generally opposed to. But there is the issue with Sansui's strange multivibrator implementation that in my opinion isn't the greatest idea at all. If someone can shed some light on why the original circuit is supposed to work without killing the transistors then please do... If I remember correctly, someone replaced these exact transistors on my original board. Coincidence? Maybe. One thing is clear to me: My replacement transistors did not like to be in this circuit and I wouldn't want to be in there either
Have two options: reverse protection diode (my current bodge for Rev A, don't like it tbh) or completely avoid this circuit and use something fancy like the 555 timer for the blinking (was available at the time if that matters). (Btw. there is not to much space on the board to work with, so no complex discrete circuits...)
Haven't decided yet, I've done two variants (KiCAD only, not actually built).
btw.: These measurements took quite some time, so I still have to fix some CI/CD stuff for the gerber exports and prepare some documentation for the driver boards...
Last edited:
Sebastian, this update is a masterclass in test bench methodology!
1. Square Wave / LTspice Sim: Seeing the LTspice simulation match the physical measurements so closely (63 kHzpredicted vs. 64 kHz measured) is fantastic. Isolating the ringing as an external LC resonance between the output inductance/cables and the load capacitance—rather than internal NFB loop instability—proves that the Rev D driver topology is rock solid.
2. IMD Sweeps: Very clean spectrum across the power range up to 58W! That confirms excellent linearity.
3. F-2663 Board & Multivibrator: You hit the nail on the head regarding that original Sansui blinking circuit. It’s a well-known pain point, and replacing the fragile discrete multivibrator with a classic 555 timer is honestly the smartest engineering move. It’s period-correct, dead simple, bulletproof, and fits the tight PCB real estate easily without needing bodged protection diodes.
Take your time with the CI/CD scripts and documentation for the driver board Gerbers. Thanks again for taking the time to run all these extra tests!
1. Square Wave / LTspice Sim: Seeing the LTspice simulation match the physical measurements so closely (63 kHzpredicted vs. 64 kHz measured) is fantastic. Isolating the ringing as an external LC resonance between the output inductance/cables and the load capacitance—rather than internal NFB loop instability—proves that the Rev D driver topology is rock solid.
2. IMD Sweeps: Very clean spectrum across the power range up to 58W! That confirms excellent linearity.
3. F-2663 Board & Multivibrator: You hit the nail on the head regarding that original Sansui blinking circuit. It’s a well-known pain point, and replacing the fragile discrete multivibrator with a classic 555 timer is honestly the smartest engineering move. It’s period-correct, dead simple, bulletproof, and fits the tight PCB real estate easily without needing bodged protection diodes.
Take your time with the CI/CD scripts and documentation for the driver board Gerbers. Thanks again for taking the time to run all these extra tests!
SebastianH
Member
Update:
Gerber files of left and right channel driver boards published on my website. The website layout is a bit buggy, I'll fix it soon...
License is CC-BY-NC-SA-4.0. No commercial use. As is.
Link to the repo
Gerber files of left and right channel driver boards published on my website. The website layout is a bit buggy, I'll fix it soon...
License is CC-BY-NC-SA-4.0. No commercial use. As is.
Link to the repo
p4spooky
Active Member
Thanks for the files. I am going to order these boards. Have a AU 717 for restoration. Will start with the driver boards. Once you release the power supply board will get that completed. Brilliant work and amazing contribution to our community of Sansui fans
SebastianH
Member
You're obviously not obliged to report back, but I'd much appreciate feedback (and pictures). This would be especially important if something doesn't work as expected, so we can fix things before someone else orders the boards.
I would prefer to host the repository on a proper platform, but I'd also prefer not to use github. Maybe gitlab? I don't know
I would prefer to host the repository on a proper platform, but I'd also prefer not to use github. Maybe gitlab? I don't know
SebastianH
Member
Power Supply Protector Board with TLC555:
Any suggestions?
Any suggestions?
I was just looking at the diagram above, and was attracted to the lower left and right corners of the board where you have 2 sets of pins labelled '02' & '06' (LHS) and '01' & '05' (RHS), these are in association with '08' & '10' (LHS) and '07' & '09' (RHS).
I can't make sense of the connections to the fuses F02/F04 & F01/F03 and the distinction between the mains input (Red) and secondary transformer (Purple) connections, as it looks like the track pattern Red connections (mains) are joined to the Purple connections (secondaries)?
Am I seeing things?
I can't make sense of the connections to the fuses F02/F04 & F01/F03 and the distinction between the mains input (Red) and secondary transformer (Purple) connections, as it looks like the track pattern Red connections (mains) are joined to the Purple connections (secondaries)?
Am I seeing things?
Last edited:
SebastianH
Member
Good point! Haven't thought about it. I used the EU connection pattern, J25, J26 not populated... Doesn't work for US/Japan models, I guess. Sansui used wires, I didn't. This decreases flexibility in favor of a nicer layout. Hmm... Probably should remove J25, J26 foot prints and the wire connections 9/10 completely to avoid this potentially dangerous confusion.
Not so sure about the US/Japan variants. Probably requires two separate versions in order to avoid wires.
Not so sure about the US/Japan variants. Probably requires two separate versions in order to avoid wires.
Last edited:
SebastianH
Member
Changed. Did a quick google search but I haven't found any picture that would indicate use of the other variant. Would be helpful if someone with the US variant could post a picture.
p4spooky
Active Member
Will definitely report back, I am sure I will need some hand holding while populating the boards. Got six projects ahead of this as this so going to be around Nov before I will be working on this. Will keep this thread updated once the boards are ordered.You're obviously not obliged to report back, but I'd much appreciate feedback (and pictures). This would be especially important if something doesn't work as expected, so we can fix things before someone else orders the boards.
I would prefer to host the repository on a proper platform, but I'd also prefer not to use github. Maybe gitlab? I don't know