Sansui AU-717: Restoration with new boards

[SebastianH]

I use LSK389B on F-2806 for a G-8000 and works very well.

Good to hear. I think in this case (2SK97 rank 1) the LSK389A should be more suitable, since the tail current is only ~3mA...

If the VAS transistor is saturating, maybe it's something to do with the values of C12 & C13 on the F-2721/F-2722 PCB, or the replacement chosen for TR06?

I replaced C12/13 with a 1uF film capacitor instead of 1uF El. cap || 47nF film. Do you think the reduced ESR/ L would have a meaningful impact there? To me, the different transistors seem to be a more likely cause. I could try the nominally worse TTC004B/TTA004B in one channel and compare the measurements. That would be quite interesting, actually. For TR06 I went with a KSC945C instead of the original 2SC945.

Are you selling your new blanks?

Nice to hear that there is interest in the PCBs! I haven't decided yet how to proceed, ideally I want to sort out the mentioned issues first - if I'm able to.

 

[SebastianH]

I'm still looking for additional measures to improve the noise situation/problem* using my existing "prototypes". In the living room the noise didn't bother me that much, but it couldn't get unnoticed in my listening spot where I planned to use this amp. So here is what I came up with so far:

1. JFETs​

Right now, I'm waiting for the LSK389A to arrive. For the original 2SK97 no official datasheet seems to be available, but thankfully, user smurfer77 posted a couple of specs here, including an input noise voltage of 13 nV/sqrt(Hz) (R_G = 10k, f=1kHz). To me, the test conditions are a bit unclear. Still, it should be safe to assume that the LSK389A (1.3nV/sqrt(Hz) typ) has much, much lower noise.

There are a couple of other factors that may contribute to the noise in a significant way - only after the 2SK97 was replaced - and I probably want to check out or discuss the following:

2. Input resistor R02​

R02 is a 2.2k resistor with a thermal noise is about 6nV/sqrt(Hz). When fed by the internal preamp one has to add ~0 to about 1.4k (depending on the volume pot position) to this => up to 8nV/sqrt(Hz).

I consider reducing R02 to 1k or even 470 ohms and possibly increasing C01 from 100pF to something like 220pF, but i fear high frequency oscillations with the PCB design that I kept close to the orginal. There are 4-5 cm (about 2") between C01 and the JFET. The track's L and the JFET's C could be a great HF oscillator I fear?!

What I could try: Cut the "input track" and place SMD R and C much closer to the Gate.

• 1k => 4.1 to 6.4nV/sqrt(Hz) of thermal noise
• 470 ohms: 2.8 to 5.6 nV/sqrt(Hz).

This would be much closer to the feedback network's thermal noise:

3. Feedback network​

(1.2k || 10k) || 27k ~ 1.03k => 4.2nV/sqrt(Hz) thermal noise. That is negligible in comparison to the noise contribution of the input in its original form. I could reduce the resistance while keeping the ratio the same. But imho it doesn't make too much sense to mess with this because of the preamp and input configuration of the power amp, even more so in unmodified form (see point 2).

4. Gain​

Like @Hipocrates hinted at (in case this amp had a really high gain, which this one doesn't), I could reduce the amplifiers closed loop gain and therefore the noise gain. But this would be more a last resort to me, because 1Vrms sensitivity seems to be a reasonable/desireable choice.

5. Constant current source (not in the original design)​

Add bypassing for the reference with 10uF SMD ceramics (size constraints with my existing design). Would give me -3dB at 0.7 Hz (10uF, 22k).

​

6. Zener power supply​

already bypassed by 47uF electrolytic (up from 33uF). Add 100nF ... 1uF ceramics for good measure, but I'm not sure it will do much for the noise within the audio band.

How to proceed then?​

My suggestion would be to try improvements in the following order:

• 1: JFETs first, because I think this might be the biggest contributor.
• 5: Probably a good idea anyway, minor modification
• 6: Probably doesn't hurt anyway, minor modification
• 2: Possibly fairly significant noise contribution as soon as the 1. is implemented
• 3: Probably doesn't help much, unless other modifications are implemented
• 4: Last resort, because this is a significant change to the operation of the amp

If you have any suggestions or corrections, please let me know.

BTW: I already use thin film and metal film resistors on the board. As far as I know cermet trim pots are pretty noisy, but the one used in the input stage is low value (VR01=100 ohm; VR02 doesn't exist on my boards/would be in parallel) anyway.

- Sebastian


* I came to believe that the amplifier in its original form just is somewhat noisy.

 

[Sui-Yogi]

From my experience and only , I have no proof or measurements to provide to support my claims but what I heard , I heard it and it cannot be unheard .

Sansui uses JFETS at the input stage , not a big deal , any audio grade, electrical with in the safe operating area , thermally and gain matched ( if are not dual ) they will do the work and most likely better than the original.
If you use singles and you can’t match the IDSS exact , you might have a bit of an offset, insignificant to cause problems .
That been said, the JFETS are not aging well and compare to the transistors are not “dying”. Out of the circuit will measure well with in their specs but under load and heat that will be a different story .
The LSK389 is a great device unlikely that even a new 2SK97 would exceed LSK specs.
Further more, I trust or use anything “made in China” in a Japanese equipment, is like the equipment knows it and does not like it.
Made in China Specs on data-sheets most likely are manipulated to stand out in the market . ( this is an assumption as I never understood how a KSC/KSA TO-126 as a
pre-driver can sound so bad with so great specs )
Regarding equivalent replacements , my experience is pointing me towards
Made in Japan , USA, EU only , worst case made in Taiwan .

Again, so I don’t get misunderstood the above is what I heard , I have not measure anything and my experience is coming from hundreds of replacements I have done and listening comparisons.

Looking forward to hear how tackled this one!

 

[SebastianH]

Thanks for the feedback! Due to the lack of alternatives I used ISC (see drivers on pictures below). I experienced my own confirmation bias etc. numerous times, so - as you probably already guessed - it's hard to convince me without measurements

That said: Your post triggered another search for alternatives, and I found some that I ordered a couple of days ago: KTC3503/KTA1381 for VAS/predrivers, KTB1369/KTD2061 as the drivers, all made by KEC, and I plan to test those. For now, I do not expect significant changes in the performance of the amplifier, but I'm happy to be proven wrong. Happens all the time, see below.

What I like about the KTB1369/KTD2061 is their full plastic TO-220IS case that will allow me to get rid of the plastic washer. I'll report my findings...

Update​

In the mean time my order arrived with the LSK389. In preparation I did some more measurements, looking at the noise specifically. I found a couple of potential issues that do not necessarily have anything to do with the JFETs etc.: There is a certain amount of hum - expected - and more importantly more harmonic content from it (or the rectification) than I expected. Also, there is some noise from 10+ kHz .. 30 kHz that seem to originate from the amp modules.

I had to try the newly arrived LSK389 anyway. So I implemented points 1, 5, 6 from the list of my previous post and took some pictures of the modules that I've forgotten to take the last time the modules where out.



The original output transistors:

The new JFETs​

Yep. I ordered LSK389A. The label on the bag says LSK389A. Why is there a C printed on my devices?





All the same, it's indeed I_DSS range C. Even reputable distributors make mistakes, I guess. Now waiting for the new delivery. It's a very good idea to check the parts before soldering

Of course I tested the amp with the new JFETs that I had soldered in. I'd say there is not a huge difference in performance. The noise is roughly the same, because it is more likely dominated by something else. For example the 10..30kHz issue that is still present with either JFET (2SK97, LSK389C). (The THD for high frequencies seems to be a bit lower, which should be consistent with a higher open loop gain - g_m should be higher.)

The 10..30 kHz problem​

At first I suspected switching noise from isolated gate drivers of the solid state relay, but disconnecting its power supply completely and shorting the FETs (i.e. "closing" the relay) manually didn't influence any of it. So my current theory is some very low amplitude HF oscillation within the amp modules procuding some mixing artifacts in the upper audio frequency band?!

But more importantly: I thought to have narrowed the main issue down to the power amp. I found this to be surprising at the time. And now I think that I was in fact completely wrong. The large amounts of noise that I was complaining about are only present with the preamp connected (either through the switch or a jumper link). The position of the volume pot does not matter. Since the fully ccw position of the volume pot should present something close to a shorted power amp input, I now have to investigate where this noise (not so much hum) is injected. It is almost like the the volume pot (the wipers after the preamp) is not even in circuit. This would completely explain my issue.

I'll look into this later.

 

[SebastianH]

Nope, the potentiometer is fine as well, wrong again...

It's just a grounding issue. To me it looks even like a design issue and I wonder whether all amps have it: The power supply return for the preamp is the shield of the fairly long audio cable, meaning power supply and signal share an inadequate conductor (or there is a bad connection somewhere, but I still measured a reasonably low resistance, so I doubt that this is it). If I momentarily short out this connection with a decent sized conductor the excessive noise is virtually gone. What a difference.

I have not run all tests yet, but brought it over to my listening room and I'm optimistic.

Disclaimer: If you do such a modification, you do this at you own risk.



Since I have to wait for the correct JFETs and new transistors to arrive (probably 1-2 weeks) I could address the actual preamp noise, i. e. with the volume pot turned up. I don't think there is too much to do though. I might replace a couple of carbon film resistors with thin film. And I could add ceramics for improved decoupling (I don't think this will make an audible difference tbh).

(Actually, I'd love to replace the preamp board as well, but I'm not sure I want to go this route Can be a lot of work to get all the dimensions right the first time.)

EDIT: After further inspection this might have been a deliberate decision. The power supply connector has an unpopulated ground pin, that could have been used. Either way, without the connection my amp performed poorly. I might try using this pin for a factory look. On the other hand, having modifications easily visible can be advantageous.

 

[Magnus Back]

If I read the schematic right the chassis ground is supposed to connect to pin 27 on the side, routed through F-2723 and F-2667.
That connects through to same PCB trace as the unconnected middle ground pin for the power connector.
Is that one connected on yours?

 

[SebastianH]

Appreciate the comment!

Yes, pin 27 is connected to F-2667, but as far as I can tell there is no DC path to chassis ground on the left hand side of the schematic, only a single AC-coupled grounding point close to the ouputs (Edit: inputs). The DC grounding point(s) are the metal parts that connect to 02 of the amp modules. If I disconnect the 3 pin connector in the center of F-2720 (red cable to F-2670), the preamp has no DC path to ground.

I would not recommend powering the amp up with the connector unplugged. I did without thinking about it and then there was abs(36V) between the preamps ground and the chassis ground. Not ideal...

But if you're working on your amp right now, without power applied, you might be able to unplug the connector and measure the resistance between the ground on the preamp (F-2720) and 02 of the amp modules. I reckon you'll measure an open circuit for a DC resistance measurement (or some capacitance for an AC measurement).

Or I missed something, then please let me know

To clearify, just in case I haven't explained it properly before, although you probably understood correctly: The preamp's and power amps noise measured individually was reasonable. Only when I connected both, I had a very high noise level even with the volume pot turned down completely. Expected behaviour would be a noise level close to the noise of the power amp.

 

[Sui-Yogi]

Just a thought :

When you made the new boards , did you follow the schematics and point to point the original board, confirming the circuit ?
Mind that the schematics are full of errors.

 

[SebastianH]

Just a thought :

When you made the new boards , did you follow the schematics and point to point the original board, confirming the circuit ?
Mind that the schematics are full of errors.

Good point. To be honest, I was rather impressed by how few errors I've found, because I heard those stories for Sansui in particular I guess this is more true for other models or maybe a matter of perspective. One error that I found first while studying the schematics I had to re-discover later: I copied the following error in the protection circuit when drawing the schematics, only to notice a discrepancy between the original layout and my netlist during the PCB design.



I used the board layouts shown in the service manual, not the actual boards, because I started even before I disassemled everything. I'm fairly confident that the original boards of my amp match the layout shown in the service manual - just did a quick visual inspection and haven't found any issues. I also verified by measurement that the ground of the regulated power supplies are not tied to ground on my power supply board.

Also, I was a bit disappointed with the noise level before the restoration (as I remembered later)... There is nothing like absolute certainty, but I don't think the new boards are to blame.

 

[Magnus Back]

Appreciate the comment!

Yes, pin 27 is connected to F-2667, but as far as I can tell there is no DC path to chassis ground on the left hand side of the schematic, only a single AC-coupled grounding point close to the ouputs (Edit: inputs). The DC grounding point(s) are the metal parts that connect to 02 of the amp modules. If I disconnect the 3 pin connector in the center of F-2720 (red cable to F-2670), the preamp has no DC path to ground.

Like this?
The incoming signal ground is then ground up to the point of the pre-main switch and from then on it's chassi/power ground.

 

[SebastianH]

What I did with the comically thick wire in the picture:

:

Marked in green solder connections, marked in red gold plated connectors.

The regulated power supply on F-2663 is referenced to chassis ground only through F-2723 (phono), F-2720 (preamp) and F2670 (pre/mains switch), yes. For me it was only about the connection between the preamp and essentially the power amp.

Any power supply current that has to return through ground will lead to a voltage drop/noise on the very line the signal is referenced to. I can't fully rule out that there is some slightly dodgy connector that contributes to the noise issue I was experiencing. The connectors are gold plated and were reseated (more than once during testing).

Here a simple simulation of the effect of 50 mohm between what is essentially chassis ground and the preamp ground for example due to long cables: The supply is not completely symmetrical (e. g. only one voltage reference), hence there always is a DC current. In addition there will be hum/noise. (Note: For my board I tweaked the component values slightly, so this simulation does not exactly represent the original design. But the point stands.)



I measured a voltage drop of about 600uV if I remember correctly. Therefore I don't think that it's a connector issue either.

If you own this amp, you could test whether you have the same issue:
0) turn off the amp
1) ground one input (e. g. aux)
2) set the switch to connected (preamp-main input connected)
3) set the grounded input as the source and the volume control fully counter-clockwise
4) turn on the amp
5) measure the noise at the speaker terminals (I was able to hear significant noise at my speakers)
6) switch the amp to direct coupled/separated
7) measure the noise at the speaker terminals (noise disappears).

The noise level should be almost identical when the volume is fully turned down.

 

[SebastianH]

Quick update. Still waiting on the JFETs, waiting for shipment since 2nd of March. I wonder if they had the same issue again (C instead of A), but caught it in time. Anyways...

The transistors arrived today and I replaced the CSC3503E/CSA1381E with KTC3503Y/KTA1381Y in the left channel (VAS, pre-driver).



The KTx transistors had a slightly lower gain; the difference between NPN and PNP in gain was similar, maybe a bit less. I had a fair amount of inconsistency when measuring THD/THD+N and I wasn't really able to pinpoint the reason. That said, for a 20 kHz signal I consistently measured distortion figures that increased with power more so than with the right channel (CSx). (Bias set to ~22mV for both). Without modifying the circuit the CSx parts seem to work better in this VAS (for now).

I replaced the driver transistors (iscsemi's 2SC2238B-Y/2SA968B-Y) with KTD2061Y/KTB1369Y. The iscsemi parts where similarily poorly matched (NPN: ~90, PNP: ~135) as the amp's original 2SD382/2SB537 transistors. In contrast, I more or less just grabbed a couple of both KTD and KTB and they had a much higher and very similar gain of about 180. These might be a better replacement.

Config 1​

VAS: KTC3503Y/KTA1381Y
pre-drivers: KTC3503Y/KTA1381Y
drivers: KTD2061Y/KTB1369Y

Config 2​

VAS: CSC3503E/CSA1381E
pre-drivers: KTC3503Y/KTA1381Y
drivers: KTD2061Y/KTB1369Y

Notes​

• Measured into 8 ohms
• Right channel is not modified, but measurements of this channel for config 1 and config 2 do not match exactly
• Line voltage variations move the clipping point
• Unfortunately I only have a couple of KSC3503D left, no E/F; and also no KSA1381

A lot more testing to do. I plan to test the following configs as well:

Config 3:

VAS: CSx
pre-drivers: CSx
drivers: KTx

Config 4:

VAS: CSx
pre-drivers: TTx004B
drivers: KTx

 

[Magnus Back]

I just started on a AU-717 but have to go through the regulated power supply first (voltages are way off so something is wrong) and some other issues before it's worth comparing.
Also lots of distortion so I need to fix that but then I'll check the ground and compare with an added ground like you did.

I do have some KSA1381ESTU. I can send two for you to test with?

Regarding pre-drivers I see you will test with TTA004B/TTC004C as well. Good, because those are readily available.

 

[Magnus Back]

I fixed the regulated power supply, the power amp boards and the pre/tone amp, it had a number of mismatched and odd transistors on it and of course some that should be replaced anyway.
I measured resistance to chassis ground on F-2720 and it's less than 0.2 Ohm. I did not yet add the additional wire.

If I now put shorting plugs on AUX and turn volume to max I get this.
I can't hear any real noise/hiss with headphones, only hum from power supply (multiples of 60Hz as seen in the picture).
The hum increases with volume and then goes down close to max.

Zero Volume


70% volume


Full volume

 

[SebastianH]

I do have some KSA1381ESTU. I can send two for you to test with?

Thank you for the kind offer! Since I'm not located in the US, shipping is likely too expensive. And my KSC3503D's have a rather low gain of just under 100 (it's just D), so I wouldn't want to use these anyway.

I measured resistance to chassis ground on F-2720 and it's less than 0.2 Ohm. I did not yet add the additional wire.

Yes, it's probably a good idea to fix your amp first. I'd recommend to measure this with a fairly high resolution meter, ideally with a low ohms range, ideally with a 4 wire measurement.

If less than 0.2 Ohms means greater than 0.1 Ohms too, and this measurement is accurate, then this would be quite a large value if you consider that I used 0.05 ohms in my simulation.

If I now put shorting plugs on AUX and turn volume to max I get this.
I can't hear any real noise/hiss with headphones, only hum from power supply (multiples of 60Hz as seen in the picture).
The hum increases with volume and then goes down close to max.

Where did you connect your QA for these measurements? Speaker out or preamp-out? Are preamp and power amp connected? (I'd guess yes, because you're using your headphones too.)

 

[SebastianH]

TLDR: I measured various configurations including the ones mentioned above. I achieved the most consistent results with my initial configuration: CSC3503E, CSA1381E (VAS, pre-driver); 2SC2238B-Y/2SA968B-Y as drivers. And I ran into issues/worse performance with most other configurations tested. A normal user might not notice. That said, I'm generally not confident in the measurements: I had very weired issues, where for example suddenly one of the channels had high THD+N. Also: New outputs seem to perform great...

Configuration 3​

Left channel:

VAS: CSx
pre-drivers: CSx
drivers: KTx

Right channel (Initial configuration):

VAS: CSx
pre-drivers: CSx
drivers: 2Sx

Diagram: 85W into 8 ohm.



Pretty low distortion at 1kHz. But: At higher power levels higher THD at 20 kHz as with my initial configuration. What is going on with the right channel though? It was almost like there was some weired interaction between the channels. I couldn't figure it out.

Other configurations​

I tried many variants, but not the full range of combinations. The main problem is: Every time I change a component the measurement setup changes slightly (cable routing etc.). This effect can be as large or even larger than the difference I'd like to measure. My setup just isn't good enough yet.

Regarding the TTx004B: I tested the transistors too and I didn't get as good results as with the initial configuration. But again: Other factors might have played a role here. I'd expect that the TTx004B "works", but will have a bit more distortion.

After multiple hours of trial and error I went back to my initial configuration, but for now, I would consider these tests inconclusive.

Btw: One of the error sources were my scripts. I finally took the time to fix some more or less obvious issues. Now, the fundamental/generator frequency is always aligned with the fft bins. I avoid all frequencies whose harmonics may interfere with the harmonics of the mains frequency.

New Output transistors​

I tried new output transistors: NJW3281G, NJW1302G. Mine have a (comparatively high) gain of ~95-100 and are very well matched. The original transistors have less gain and are not as well matched (2SC1116: ~45, 2SA747: ~63)

The results (with my fixed scripts):

Left channel: NJWxxxxG
Right channel: 2SC1116, 2SA747

THD vs frequency:​

1W into 8 ohm:


85W into 8 ohm:


Quite a bit less distortion at high frequencies. It might be both: better matching causes less distortion, higher open loop gain makes the feedback more effective.

THD vs. power:​

At 1kHz there is no meaningful difference.



Here the difference becomes quite obvious.

I think I'll replace the outputs, seems to me like a decent improvement. Have to do more tests though.

 

[Magnus Back]

Thank you for the kind offer! Since I'm not located in the US, shipping is likely too expensive. And my KSC3503D's have a rather low gain of just under 100 (it's just D), so I wouldn't want to use these anyway.


Yes, it's probably a good idea to fix your amp first. I'd recommend to measure this with a fairly high resolution meter, ideally with a low ohms range, ideally with a 4 wire measurement.

If less than 0.2 Ohms means greater than 0.1 Ohms too, and this measurement is accurate, then this would be quite a large value if you consider that I used 0.05 ohms in my simulation.



Where did you connect your QA for these measurements? Speaker out or preamp-out? Are preamp and power amp connected? (I'd guess yes, because you're using your headphones too.)

QA is connected to speaker out. Yes pre and main are connected.
The resistance is ~0.15 Ohm with a 4 wire kelvin meter and ~0.06 Ohm with the additional ground wire connected.

 

[SebastianH]

The resistance is ~0.15 Ohm with a 4 wire kelvin meter and ~0.06 Ohm with the additional ground wire connected.

Interesting. Then I guess that's how these amps are built and you might actually profit from the additional wire too.

QA is connected to speaker out. Yes pre and main are connected.

My simple thoughts on that:

It's fairly difficult to take valid low noise measurements, so for now I'm more concerned with the relative height. (You might be a pro and those measurements are perfect, who knows .)

It's interesting that the hum decreases if the power amp is driven by the preamp with low impedance (max volume). If the volume potentiometer is turned fully ccw, the power amp's input is/should be shorted. I'd try to separate pre-amp and amp, short the power amp inputs directly and observe, whether there is any change. If the hum is gone/significantly reduced, the additional grounding wire might help here too. Otherwise I'd be looking closer at the power amp...

 

[SebastianH]

Ok. I changed the power transistors on the right channel as well. The NJWxG should be much more robust as well...



Installed the NJWxG and now it looks like this:

THD vs. frequency:​

85W into 8 ohm:



Interestingly, the results for the right channel differ a bit from the left channel: slightly lower distortion from ~100Hz to ~1kHz, higher distortion for higher frequencies. The results are a bit better than with the original transistors, but not by much.

THD vs. power​

The difference might (partially) come from the wiring inside the amplifier. Sansui used a lot of AWG 18. This is "relevant" (i. e. makes a measurable difference), especially because the positive speaker wires are long: Power Amp -> Protection Circuit Board/Relay -> Speaker switch -> Speaker Terminal. In my quick test, adding parallel wires (speaker ground and positive) did reduce the THD(+N) notably.

One of the better options to avoid this long roundtrip would be to use two relays close(r) to the speaker terminals that are only controlled by the speaker switch. Had I thought about it when designing the new protection board, I might have been able to implement those on the board. Space is quite limited on the board if it shall remain THT. For now I'm considering whether I should replace some of the wires with AWG 14 and to leave it at that. (A previous tech damaged the isolation of one of the speaker wires, so this could be a good excuse to do it.)

Because someone might think I'm crazy to go to the trouble of all of this : I don't think I will hear ANY difference between 0.01% THD and 0.005% THD or something - my speakers, my room etc. are much more limiting than the amps THD. It's just for fun. An exception would be a reduction in the noise floor that I'd likely notice (like with the mod described above, which made a huge difference).

I have to check stability with the new outputs. And maybe mouser decides to finally ship my LSK389A's...

 

[Magnus Back]

My simple thoughts on that:

It's fairly difficult to take valid low noise measurements, so for now I'm more concerned with the relative height. (You might be a pro and those measurements are perfect, who knows .)

It's interesting that the hum decreases if the power amp is driven by the preamp with low impedance (max volume). If the volume potentiometer is turned fully ccw, the power amp's input is/should be shorted. I'd try to separate pre-amp and amp, short the power amp inputs directly and observe, whether there is any change. If the hum is gone/significantly reduced, the additional grounding wire might help here too. Otherwise I'd be looking closer at the power amp...

These were quick&dirty measurements with cables all over, and settings not optimal. I just wanted to see ballpark what it looked like.
I do have one more AU-717 which I will measure as well.