Sansui AU-717: Restoration with new boards

[SebastianH]

Hey all,

after having completed a thorough restoration of my Technics SE-9600, now a restoration of my Sansui AU-717 was in order.



I got this amp in working condition, although with a DC offset of ~50mV, a slightly dirty speaker switch and - only detectable with test equipment - a slightly worse performing left channel.
Someone must have spilled some type of a liquid all over the unit - looked worse in person. Someone had done some maintenance before and replaced some capacitors and resistors, my guess is in the more recent history. They possibly missed the fusible resistors in both the Head Amp and Tone Circuit Boards that were - of course - widely out of spec.



Since replacing a couple of components for maintenance can get somewhat boring, I decided to design three new boards:

• F-2663: Power Supply & Protector Circuit Board
• F-2721: Driver Circuit Board (L-CH)
• F-2722: Driver Circuit Board (R-CH)

Spoiler: All three boards installed:



Actually, one of the reasons was the smell of the amp, once again. Some people surely like the smell of old electronics, I just don't. That's my excuse anyway.
For the most part I implemented Sansui's original circuit design. A couple of notes though:

• I generally favored available modern replacements, where reasonable.
• Tweaked some of the component values (mostly larger capacitance values for power supply bypass capacitors).
• Use of thermal pads for the thermal interfaces of all transistors; thermal compound only for the interfaces between heatsink parts
• F-2663:
  • 2x LM4040 (10V) Precision shunt regulator instead of the 13V Zener diodes in the voltage reference. Totally unnecessary, but functionally it's almost a drop-in replacement except for the resistance values in the feedback voltage divider which I had to re-calculate anyway (see next point). And yes, the voltages seem to be very, very stable - not a lot of thermal drift. Rev B with footprints for both the Zener diode and LM4040.
  • Voltage adjustment for pos./neg. regulated supplies with multi-turn pots. Also just for fun. Now both rails of both channels are pretty much exactly +/-33V. Yay
  • Use of my solid state relay that I presented here, with a minor additions to the protection circuit (can be easily configured for mechanical relays). For this application with even more robust MOSFETs (IPP022N12NM6AKSA1). This actually reduces the power consumption (and the heat produced by this board) noticeably.
  • I increased the resistance of the bleeder resistors. In the original design the resistors get pretty hot and waste a ton of power.
  • Used two parallel 3W resistors for the LEDs (2x) to at least drop their surface temperature.
• F-2721/F-2722:
  • Two-transistor constant current sinks instead of tail resistors for the JFET diff pairs. Footprint for tail resistor available though.
  • Footprints for both the original 2SK97 (DIP-6) and LSK389(A) (SOIC-8)
  • One multi-turn potentiometer for DC offset, instead of a single-turn coarse/fine adjustment
  • Heat sinks for the VAS transistors (get quite hot in the original design) and pre-drivers (probably not really required at all).
  • Driver transistors do NOT use the tab, instead the collector lead is used. Would require modification to use the original drivers with the cut collector lead. I do not like Sansui's approach from a mechanical standpoint, although it simplified the board layout, so it made sense back then. I do not have to worry about that as much with two layers available.
  • Re-used the original 2SK97, inductors, driver heat sink
  • Footprints for two different options of emitter resistors (original, BPR58 style)
  • Rev B with different mounting options for R10, so that both R10 and R04 could probably be mounted on the bottom layer side. Those resistors (3W metal film) get fairly hot (about 80°C for R10) and help those poor (105°C) electrolytic capacitors C02, C04, C11, C14 to dry out faster (measured something like 55°C for C04, so wasn't a huge deal).

Some pictures of the Power Supply & Protector Circuit Board (note that all pictures show the first revision of the respective boards that I actually use; I documented required changes in a second revision, but have not tested the changes yet):







I somehow managed to not take a single photo of either amp module before installing 'em. So these pictures have to suffice:





A couple more:





I should be able to provide some before and after measurements soon; I hope for similar (and not necessarily better) results as with the original, given that I replaced all semiconductors, mostly with different types.

If you have any questions, I'm happy to answer.

Sebastian

 

[dcmeigs]

Impressive! I’m guessing this wasn’t your first rodeo with board design.

 

[SebastianH]

Nope

 

[The Fuxtor]

Love it. Very nice work!!

 

[jmjrcv]

I'm impressed with the effort as well, particularly since I went through a similar board design effort for a broken (my own fault) F-2721 board in my AU-517. Are you willing to share the schematic and/or BOM for the amp boards? My goal is to make a Frankenstein AU-417 using the AU-517 amp boards but I've gotten distracted from building the boards and have not even considered what modern parts might best be used. KiCad is my new friend

 

[Sui-Yogi]

Very nice work !

I should be able to provide some before and after measurements soon; I hope for similar (and not necessarily better) results as with the original, given that I replaced all semiconductors, mostly with different types.

I’m interested to hear what you heard and measure before-after .
The AU-717 is a very popular model around Sansui enthusiast.

Use of my solid state relay that I presented here, with a minor additions to the protection circuit (can be easily configured for mechanical relays). For this application with even more robust MOSFETs (IPP022N12NM6AKSA1). This actually reduces the power consumption (and the heat produced by this board) noticeably.

The signal with mechanical relay is traveling through the contacts , pretty straight forward.
In favour of heat and power reduction
( if I understand well the function of the solid relay) you pass the signal through the MOSFETS ?

 

[Hipocrates]

I should be able to provide some before and after measurements soon

Fantastic work!!

 

[SebastianH]

Thanks!

Are you willing to share the schematic and/or BOM for the amp boards? My goal is to make a Frankenstein AU-417 using the AU-517 amp boards but I've gotten distracted from building the boards and have not even considered what modern parts might best be used. KiCad is my new friend

This is the kind of project I love to hear more about I consider publishing it on github once everything is ready (as is, likely under a license that doesn't allow commercial use and forces you to publish your changes/copy left, because then I want all others to profit from those changes too).

Modern is relative; I should have said: "more modern". For example, I used a variant of the good old 2SC3503/2SA1381 for the VAS and as pre-drivers and 2SC2238, 2SA968 as the drivers. Most of those are likely obsolete too. I tried TTC004B/TTA004B in my Sansui AU-317 in places where the 2SC3503/2SA1381 likely would have been the best match (especially C_ob), without any real issue. Actually, this little amp performs great and is one of my favorites. Might post a short report on this restoration too.

I forgot to mention that all my output transistors (2SC1116, 2SA747) are still fine and these (and the LEDs!) should be the only original semiconductors left. I wouldn't consider MJ21193/4G to be a true alternative and would try NJW3281G/1302G (TO-3P) first, if I had to.

I’m interested to hear what you heard and measure before-after .

I will happily share the measurements (maybe I start measuring later today), but unfortunately I won't be able to give you a meaningful A to B comparison of the sound: This amp took me many, many hours over a period of a couple of months during which I did not listen to the amp - designing, ordering, assembling, bench testing the individual PCBs and finally the actual restoration. So at this point I can't remember how the amp sounded before. Now consider, that the amp uses the same circuit design and likely/hopefully performs similar (i. e. similar specs)...

What I will say: Subjectively the amp sounds great (listening to it while writing this post). But this doesn't mean much to someone else's ears

The signal with mechanical relay is traveling through the contacts , pretty straight forward.
In favour of heat and power reduction
( if I understand well the function of the solid relay) you pass the signal through the MOSFETS ?

Yes, each channel uses two MOSFETs in series (common source) with the positive speaker output of the respective amp module. The actual schematic of the MOSFET-PCB of the solid state relay looks like that:


The MOSFETs used have a R_DS(on,max) of just 2.2 mohm each, and an almost crazy high single pulse avalanche energy of about 1.5J. With the additional track and pin resistance I would expect something like 6-8 mohm, which is as good or even better as good mechanical relays in a similar package, but without any wear...

What I was referring to when mentioning the lower power dissipation was not the usually rather small power dissipation in the signal path, but the quite substantial dissipation in both the dropper resistors (to drop about 55V idle down to about 24V for the relay) and the relay coil itself. This combined can result in about 1.5W of power dissipation for the relay drive (about 30mA). The solid state relay's drive circuits consume about 5mA, hence the power dissipation drops to something like 0.25W.

So solid state relays have similar to lower on-resistance, are much more robust in case of fault conditions (if properly designed!) and waste less power in the drive circuitry. On the other hand you can get more poles/throws in electro-mechanical relays than with my design (which doesn't matter in this particular application), you have an audible feedback of what the protection circuit is doing - which can be very nice - and my solid state relay is probably 2 to 3 times as expensive as a typical omron MY2/MY4 relay. Ah, and if you're afraid of soldering tiny SMD parts - 0603 and the likes - than just use a conventional relay...

 

[Sui-Yogi]

@SebastianH thank you for the explanation above , much appreciated!
Looking forward for the measurements too, indeed “how it sounds is very subjective”.
I have no problem soldering 0603 case , even that I think I’m reaching my limitations in that size/case .
My concern about the solid relay ,,was more about traveling the signal yet through an other circuit , when the “goal” is to reduce the signal path ?

 

[SebastianH]

I have no problem soldering 0603 case , even that I think I’m reaching my limitations in that size/case .
My concern about the solid relay ,,was more about traveling the signal yet through an other circuit , when the “goal” is to reduce the signal path ?

For anything smaller than 0603 I absolutely need the microscope, but I don't solder 0402 parts willingly

In this case I'm not too worried about that. Solid state relays do not necessarily introduce much distortion at all, as an example see for example Neurochrome's Guardian-86 protection circuit and the measurements. This surely doesn't mean that my relay performs as well, but I haven't characterized my relay like that. Also, I do not own an APx555B audio analyzer and an ultra low distortion amplifier...

That being said, there are surely modes of how a solid state relay may introduce distortion. For example, if the gate driver modulates the gate voltage, this could change the drain-source resistance ever so slightly, leading to distortion. On the other hand: Many relays have a higher contact resistance to begin with and they certainly degrade over time, leading to substantial distortion...

 

[SebastianH]

ok, I got around to taking some preamp measurements. These are potentially influenced by the following actions:

• New Power Supply & Protection Circuit Board
• Changes on the original Tone Control Circuit Board:
  • Replacement of electrolytic capacitors
  • Replacement of out-of-spec fusible resistors
  • Replacement of VD-1212 double diodes (1N4148). Btw.: I measured three of the VD-1212 diodes and one of them was open. Since I haven't noticed any issue with the tone circuits and briefly checked the frequency response of the phono stage and everything seemed to be fine before, this diode may have been destroyed from merely desoldering it. I guess it was a good thing to replace those diodes...
  • Replacement of all transistors (matched KSA992FB for diff pair and constant current source; KSC1845F, KSA992FA reasonably well matched)
• Minor impact from changes (like new bypass capacitors) on the Equalizer Circuit Board (that features the input selector switch, but just passive for line inputs).
• routing cables slightly differently and polishing the RCA connectors could have a certain impact as well. I do not think that this influences the results by much.

A couple of other points:

• It is fairly simple to make seemingly small changes to the measurement setup/mistakes that could have a certain impact on the results. This is especially relevant, when comparing results before/after with the measurements taken exactly two month apart... Just two examples:
  • Power supply via isolation transformer vs. a different outlet at my bench. This directly influences the noise floor/hum. (Maybe I should investigate...)
  • measuring THD/THD+N at slightly different input/output levels
• I'm personally not that interested in record players - I do not own one. So I won't show measurements of the "head amp".

Measurements taken with QA403.

(Power supply) hum​

Please ignore the title of the diagram: It's averaged, so it's not a correct representation of the noise, it shows the amplitude of the 50 Hz humm and its harmonics.

Gain ~0 dB

Before:



After (isolation transformer):



After (different outlet):

	before	before	after (iso)	after (iso)	after (outlet)	after (outlet)
f / Hz	left (dBV)	right (dBV)	left (dBV)	right (dBV)	left (dBV)	right (dBV)
49.99	-104.17	-103.54	-108.49	-106.23	-117.10	-111.83
99.98	-120.73	-121.94	-122.01	-121.01	-128.18	-127.86
149.96	-106.08	-104.66	-107.59	-107.09	-108.85	-108.37

THD vs. frequency​

Gain ~0 dB, input 1 Vpp

Before:


After (outlet):



Very similar results.

THD+N vs. frequency:​

Gain ~0 dB, input 1 Vpp

Before:



After (outlet):



Virtually identical results (probably measurement artifacts at 10/20Hz)... The results improve drastically with higher amplitudes (not really surprising)... The noise completely dominates the THD+N figure.

Conclusion​

I also looked at IMD and many more measurements regarding the frequency response. Those are too similar to waste any more time... I do think that the power supply/new capacitors might have slightly reduced the hum. Other than that nothing really to report. Success, I'd say. Should really look into the increased noise floor when the amp is powered via the isolation transformer...

 

[SebastianH]

The power amp measurements are in.

These are mainly influenced by the following actions:

• New Driver Circuit Boards obviously
• New main filter capacitors and new film bypass capacitors - just for fun. There is likely not much much difference to the "old" new/replacement caps that the previous owner/technician must have installed - I used exactly the same KEMET capacitors again. Great specs, good physical dimension for this application, solder terminals. The only reason I changed these was a slight physical damage of one of the caps. I'll keep these and surely use at least two of them for a different project. Just in case someone has noticed: I ordered four new caps from mouser and got two different heat shrink styles. Similar (but not identical) mfr date etc. Wasn't great for my OCD

I performed some (not necessarily all) of the time domain measurements before/during the restoration, but probably have not saved most of the files. Oh well... So here we go with the measurements of the new boards only.

Measured with Siglent SDS 2x04 plus.

Clipping: 1 kHz, 8 ohm​

Looks good so far.

Clipping: 10 kHz, 8 ohm​

Some amps really do this better I don't think the original PCB/silicon showed this behaviour - or at least not as pronounced. Would be nice to repeat the measurement with the original silicon (I'd say slight dips and peaks are not uncommon, this is a bit more than I like though). The VAS transistor goes into saturation and has seemingly a hard time getting out of saturation. If someone has a specific suggestion that doesn't compromise the other performance characteristics, feel free to let me know

Step response: 1 kHz, 8 ohm, left channel (right channel identical)​

Step response: 1 kHz, 1uF || 8 ohm, left channel (right channel identical)​

Expected ringing. All good, I'd say.

Step response: Close to clipping, 8 ohm, rising edge​

Respectable slew rate of 29.3 V/us (left channel) and 32 V/us. As far as I can remember this is very close to what I've seen with the old modules.

Step response: Close to clipping, 8 ohm, falling edge​

Similar fall times.

Frequency domain

Frequency response: 1 W, 8 ohm​

Bandwidth: ~302 kHz (left), ~305 kHz (right)
Flatness (20 Hz to 20 kHz) abs(max-min): 0.086 dBV (left); 0.075 dBV (right)
Channel gain mismatch (20 Hz to 20 kHz): <=0.058 dB

Again and again very similar results. Except for the matching between the channels: matched 1% feedback resistors made all the difference. I measured my Sansui AU-317 at about 400 kHz bandwidth, not that it would matter.

 

[SebastianH]

Measured with QA403.

THD vs. frequency: 1 W, 8 Ohm​

before:



after:

THD vs. frequency: 85 W, 8 ohm​

before:



Left channel had 50-70mV DC offset. Either there was something wrong with the driver module or there was a some connection problem (dirty switch, relay, speaker terminal, dummy load, etc.)

After:

THD+N vs. frequency: 1W, 8 ohm​

Before: (would ignore <= 20Hz)



After:



Almost identical, less harmonic distortion at higher frequencies, therefore also less THD+N.

THD vs. Power: 1 kHz, 8 ohm​

Didn't measure before...

After:

THD+N vs. Power: 1 kHz, 8 ohm​

After:

THD vs. Power: 20 kHz, 8 ohm​

After:

THD+N vs. Power: 20 kHz, 8 ohm​

After:

 

[SebastianH]

IMD CCIF/ITU-R: 19 kHz + 20 kHz @ ca. 19 dBV, 8 ohm​

Before:



Yeah, the left channel. Whatever...

After:

IMD SMPTE: 60 Hz @ ca. 19 dBV + 7 kHz (4:1), 8 ohm​

Before:



After:



There is more, but I guess, this is way more than enough...

Conclusion:​

The new amp modules perform very similarly in most tests:

• The previously mentioned clipping issue at higher frequencies is still present.
• Very similar step response.
• Generally similar THD at low and mid frequencies, with lower THD at high frequencies.
• Similar THD+N compared to before.
• Similar IMD performance.

Overall, I can’t really complain about the results. That said:

Today I put the amp back into my listening room. Apart from the overall great sound, I immediately noticed a hissing noise that I hadn’t noticed as much at the workbench (same speakers, but probably the worst placement imaginable for listening...).
Interestingly, my Sansui AU-317 subjectively seems to have less hiss, although it does have a hint of hum (I couldn’t hear any hum from the 717). However, the AU-317 measures significantly worse (THD+N). I don’t fully remember whether the hiss was already there before the restoration, but the measurements would suggest that it was.

On the other hand, my measurement setup (cables, dummy load, etc.) has improved somewhat since then. So it’s possible that the new modules had a slightly easier time achieving the same or better values. Or maybe not — I honestly don’t know. It was also quite easy to get worse THD+N results simply by changing the amp position or cable routing.

One of the components that likely has a major impact on the overall noise floor — the dual JFET input pair — was left unchanged (2SK97, grade “1” if I read it correctly, i.e. the lowest I_DSS grade).

Out of curiosity, I tested the relatively inexpensive JFE2140, but:

• Its I_DSS is significantly different from the 2SK97.
• It showed surprisingly high leakage current, which threw off the DC balance when switching between AC-coupled and DC-coupled inputs.

My (possibly completely wrong) theory is that this could be related to the built-in ESD protection circuitry — even though in my layout both relevant pins are left floating, as suggested in the datasheet. Does anyone with JFET experience have an explanation for this?

The JFE2140 seemed much more temperature-stable and - in this operating point - appeared to produce slightly lower distortion. I also had the impression that it was less noisy, although, I didn’t perform proper noise measurements to confirm that.

What do you think? Should I try the expensive LSK389A? On paper it looks like it could be a very good (low noise) match.

 

[Hipocrates]

What do you think?

How much gain this amp has? I've seen many vintage ones with 36dB or so.... maybe lowering the gain helps?

 

[SebastianH]

28.3 dB (1Vrms -> rated power 85W). Seems reasonable to me.

I think I'll give the LSK389A a shot.

 

[Dswankey]

Excellent!!!, what a great amp they deserve to be restored.

 

[Hipocrates]

Seems reasonable to me.

Yupe.

I think I'll give the LSK389A a shot.

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

 

[Hyperion]

If someone has a specific suggestion that doesn't compromise the other performance characteristics, feel free to let me know

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?

 

[Plimpington2]

28.3 dB (1Vrms -> rated power 85W). Seems reasonable to me.

I think I'll give the LSK389A a shot.

Are you selling your new blanks?