Tips for success with re-building your GFA-565 circuit board… in no particular order…

Be prepared for a lot of work! It’s more than people expect. But it is very satisfying, like a big puzzle, and the result is a truly spectacular amplifier.

This is an advanced electronics project, so if you haven’t had much experience fixing amplifiers, the GFA-565 is absolutely not a good place to start. It is a fairly complicated beast.

You’ll need a fairly well-equipped electronics laboratory, with an oscilloscope, signal generator, and preferably a distortion meter. A variac is pretty much essential for powering the amp up the first time.

I want as many people to succeed with their repairs, and I am happy to answer questions about the board itself, to clear up any confusion about what parts go where and such, but if you need help actually troubleshooting the amplifier, I need to charge my usual hourly rate($60/hr) Please read up as much as you can on the repair of the GFA-565. There are many edifying discussion threads on the topic at DIYAUDIO.COMThis one and this one in particular. It’s a lot of reading, but you’ll find important tips that may save you time and hassle.

Go slow, it’s faster.

Here’s the latest board revision!

They are two-layer boards, though the layout is still almost exactly as original. (No need to mess with a known working layout.) The two layers allow me to add a few new features: All the small wire jumpers are now built in as a trace on the top layer. The two longer black wire jumpers are still required. There are now pads to surface-mount CMXSTB400 stabistor diodes in place of the original KB262 and KB362.


Parts: Here’s a spreadsheet with part numbers from Mouser for all known equivalents. Quantities are exactly enough to populate one board. Order extras if you need to match transistors or resistors. You don’t need to replace everything on the board, but I recommend at least replacing all resistors and small signal diodes. (except for the stabistors, keep those if yours are OK. They are delicate, so be careful de-soldering.)

Resistors listed are all 0.1% tolerance in symmetrical applications, and 1% elsewhere. Dale RN55C is my preference, but they are not available in all resistance values, so TE Connectivity YR1 0.1% series resistors are used where Dales cannot be had. You may find that stocks have changed since this writing.
The TO-92 voltage references labeled “Adcom J2” are often ruined, being located right next to the leaky caps. I usually replace all four. The other voltage reference on the board is usually fine. (The 2.5V Adcom J6 or LM336)

Notes: (In no particular order)

  • Error in the service manual: In the parts list, the first mention of R114 should read R144.
    R144, R145 1/4w/499ohms
    R114, R145 1/4w/499ohms
  • Reusing original parts: Be cautious about re-using the original components. The bad capacitors spray electrolyte all over the board, and a drop of spittle can create a conductive path between component leads, even on the undersides of transistors. Clean between the leads with a sharp cotton swab. All re-used parts should be run through an ultrasonic cleaner. For a solvent, I use a 50/50 mix of denatured alcohol or vodka, and Simple Green. Rinse the parts in water afterwards and dry thoroughly. Your nose can tell you if there is still electrolyte remaining on a component lead. Just heat the component lead with a soldering iron and sniff. The smell is very distinctive; a bit like rotten antifreeze. Once you smell it, you’ll never forget it. To get an idea of what it smells like, try heating up a component pad around one of the bad caps on the original board. Gross, huh?
  • STABISTORS: D105 through D108 (KB262 and KB362) are listed incorrectly in the service manual as “Varistor Diodes”. They are actually stabistors; a type of diode that has an especially small change in forward voltage drop versus current—about half as much as a normal diode like a 1N4148, giving a steeper curve, which is good for regulation. Each stabistor diode is actually a package of stacked diode junctions in series, with approximately 0.6V forward drop per diode. They are named for the number of junctions in the package—KB262 has two diodes (1.2Vf), KB362 has three (1.8Vf), etc. There are no modern replacements available in through-hole packages. However, (after much research) I found surface-mount versions are still available from Central Semiconductor. The CMXSTB400 has four stabistors in one package, and the circuit requires two diodes with two junctions, and two diodes with three junctions, so my board uses the CMXSTB400 to replace all four stabistors. (The extra diodes simply go unused.)
    You can test stabistors with the diode check function on a multimeter. They should read around 1.2v for KB262 and 1.8v for KB362. There should be no reading in reverse. On some earlier versions of the boards, it can be a little hard to tell where pin 1 is on the stabistors. These photos should help.

    Use the CMXSTB400 to replace both KB262 and KB362. The extra one or two diodes in the chip simply go unused.

    But you can just re-use the original stabistors if they are good. They are seldom a problem. They are physically fragile, so be careful desoldering. Also, be careful about the markings on the package; they are opposite what you might expect. The KB262 and KB362 diodes that come in the epoxy-blob package—unusually—have a black stripe to mark the ANODE not CATHODE. Don’t take my word on this. Confirm polarity with your meter’s diode test setting. More on this topic here at DIYAUDIO.

  • The original heatsinks run really hot. I recommend swapping them with the larger ones provided in the parts list. It’s the closest fit I could find, but unfortunately, the transistor mounting hole is higher than on the original heatsink, so the leads on the original transistors wouldn’t reach. Also, the supplied 6-32 screw hole is too large for the TO-126 transistor. You’ll need to tap a new M3/0.5mm hole at the same height as the original heatsink. Countersink the hole after tapping.
  • Matching transistors: If you replace the MPSA13 and MPSA63 devices, they must be matched. I recommend it anyways, as the factory matches are only so-so. This is an involved topic, and is covered here at this thread on DIYAUDIO. The DIYAUDIO transistor matching jig is a bit of a community effort; The forum moderator ‘Anatech’ designed the circuit, and user ‘Cogeniac’ designed the original circuit board, and I’ve developed my own forked version of that board. At some point there will probably be a group buy for boards. The project will also likely go open-source.
    Ideally, the cascode transistors Q103, Q104, Q107, Q108 should also be matched. The Fairchild KSP42 and KSP92 are confirmed to work beautifully. YES, I sell matched sets of input transistors for $50. This is a complete set of 8 matched transistors, enough for one amp. 2x MPSA13, 2x MPSA63, 2x KSP42, 2x KSP92.
  • EBC, CBE, BCE confusion! Replacement transistors KSP42 and KSP92 are EBC instead of BCE as original 2SC3478 and 2SA1376, so they must have their leads bent goofy in order to fit. This can be surprisingly difficult to keep straight in your head. Triple-check that you have them in the correct order. To make things even more confusing, transistors BC550 and BC560 are CBE instead of BCE, the opposite arrangement versus the KSP42/92. And some datasheets show pin numbers from above, some below.
    The best way to be sure of your Es and Bs and Cs, is to use a transistor tester to verify the leads are in the right order. I use the one built into a cheap multimeter. It’s not a good transistor tester, but when the leads are in the correct order, you will read a gain over 100. Compare to the original transistor outline and install.
    Here’s a screen-shot of Q103, 107, 104 and 108, showing where the transistor leads are supposed to go. (This is printed on the latest version of the boards I sell.)
    Qs103and107    Qs104and108
  • I like to get rid of the wire-wrapping posts and solder the wires directly in. When I put the amp back together, I install the output modules last, so no need for the wire-wrapping posts.
  • It’s a good idea to refurbish the soft-start board while you’re at it. The original 25W 4.7R in-rush protection wire-wound resistor often burns out. I replace it with an aluminum-cased 50W resistor. IMPORTANT: The original resistor is held in place mechanically as well as by solder. In the event of a melt-down, it should not collapse and short to the chassis. The replacement resistor should be mounted in such a way that it will not fall through or collapse if it melts. I use 12ga solid copper wire arranged as in photo below. You could also mount it to the chassis and run short wires to it.

    (Ignore the fact that the photo above shows an 8-ohm resistor, not 4.7, it’s for a special project)
  • Also recommended is to replace R501 3.3K with a higher-wattage 1/2w resistor. The original runs hot enough to turn brown.
  • Don’t forget to short out the 4.7R soft-start resistor if you are bringing the amp up on a variac. If you don’t have a variac, at least use a DBT. (Dim bulb tester) But I really, really recommend using a variac.
  • Speaking of variac testing, the amp’s bias circuit will not engage until you hit about 40VAC input.
  • The OEM board headers and plugs for the BIAS and BIAS COMP connections are kind of crappy, and the plastic gets brittle eventually. I replace them with the Molex Nano-Fit system. You need to buy three parts; Headers, Receptacles and Pins. Part numbers are in the spreadsheet. Assembly and crimping of the wires to the pins is supposed to be done by robot, but if you have keen eyes, a fine-tipped soldering iron, and a steady hand, you should be able to do it. The Nanofit headers have a tiny plastic locating pin on the bottom that should be clipped off so the header sits flush with the board.
  • If you replace the 2SC3298/1306 TO-220 drivers with On Semi MJE15032/33, you may have an oscillation issue, which is solved by also replacing the 2SC3907/1516 that forms the second stage of the darlington, with On Semi NJW1302/3281. I do recommend replacing these as a matter of course, as the breakdown voltage headroom is improved.
  • The output zobel network’s resistor and capacitor can be mounted directly to the binding posts instead of on the input board. I think this is much better than the ten inches of wire used to run it back to the input board as original. There’s no reason for it to be on the board; it connects to nothing there. Part numbers for a nice polypropylene cap and non-inductive resistor are in the spreadsheet.

Testing the board:

I recommend testing the board before you hook up the output section, and risk blowing out twenty expensive output transistors.

The best way to test the board is outside of the amp, hooked up to a lab power supply.

Conditions required to operate the board without the rest of the amp connected:

  1. Override the bias delay: The bias delay consists of an opto-coupler and switching transistor located on the power supply relay board, whose job it is to turn on the amp’s bias circuit after a short delay, in order to mute the amp while it powers up and stabilizes. This is the purpose of the long wire that goes from the power supply board to J106 on the input board. We’ll simply bypass all that. All that’s required is a jumper from the center pin of J106 to point 17, and that will energize the current source circuit.
  2. The thermal protect LED does not need to be connected. It doesn’t hurt though.
  3. The thermal fuses do not need to be connected. It doesn’t hurt though.
  4. Output and feedback hack: Solder a pair of standard 1/4W 1K resistors into points 3 and 4 and tie them together on the other end. (Y-Connection) This is your output. From there, run a short wire to the feedback input at point 7.
    GFA-565 Output and Feedback hack
  5. Thermal-compensation bias transistors: These are the small 2SC3478 and 2SA1376 transistors that are mounted to the output heatsinks, and connected with long wires to J104 and J105. They’ll need to be connected while you test, so just set the output module heatsinks on your bench near the board. (With nothing else connected to the output modules.)
  6. Power supply requirements: The board has two negative and two positive voltage inputs. (Two fused positive and negative supply lines feed the current source circuitry, and two un-fused positive and negative supply lines feed the amp circuit.) These can simply be wired together in parallel for the test. Power supply ground connects to point 18, +20V to points 10 and 5, and -20V to points 11 and 6. These points are conveniently labeled on the board as Fz B+, FZ B-, UnFz B+ and UnFZ B-.
    Configure your lab power supply as a +/-20V bi-polar supply with a common ground. Positive and negative 20V is plenty to operate the circuit. Current limit to 1A to prevent major damage if something is wrong. (1A is probably more than it consumes; I haven’t actually measured it, so please let me know if you find out.)
    If your power supply allows you to slowly ramp up voltage, that’s a good idea.
  7. Connect your test signal generator to the input on J103. Ground is the pin on the right.
  8. Connect your scope to the Y-point of the two 1K resistors you’ve hung off the drive outputs.
  9. Powering up: Set your signal generator at 1KHz, 100mVrms. Power up, and hope that none of the magic smoke escapes. If you’ve got it right, you’ll be looking at a nice clean sine wave on the scope. Distortion will be higher than the final amp, but it should be less than 1%. You shouldn’t see any wave deformation on the scope. DC offset should be less than 100mV, probably a lot less, but the circuit is operating at a fraction of it’s operating voltage, so everything will be a bit off. If the board seems to be working well, try a higher supply voltage, up to the full operating voltage of +/-85V, if your supply goes that high. Don’t get shocked. Check the voltage on pin 6, the output of the DC servo op-amp. (To avoid probe-slippage disaster, probe instead at the left-hand side of R116) If the amp were theoretically perfectly balanced, you would read 0V here. If you read something close to 15V, positive or negative, that means the servo is trying as hard as it can to correct some DC imbalance, and that means something is wrong. Once the amp is fully completed, you should read less than 6V at the servo output if things are well balanced. If not, you might have poor matching in the input transistors, MPSA13 and A63. Though, if the servo is pinned at 15V, it’s probably some other fault, not just bad matching. If all seems well, you can assemble the rest of the amp with some confidence. I suggest testing each and every component on the output modules first. Good luck! Please let me know if you have any suggestions or corrections.