Adcom GFA-535 owners; Has it always bugged you, how the right channel of your amp buzzes through your speakers more than the left channel? Yeah, me too. The buzz is a buzz-kill… so I killed the buzz! With this thing I made.
Now, the GFA-535 is a legendary little amplifier, deserving plenty of buzz, but it’s too bad so much of it comes out the speakers. Oddly, the GFA-535’s bigger sibling—the GFA-545—using the exact same amplifier boards as the 535, is nearly silent as a stone. What’s going on?
The sound of non-silence: This is the noise that comes out of the right side of a GFA-535 with the input jacks shorted. The recording was made with a portable digital recorder—an old M-Audio MicroTrack 24/96—with its microphone input wired directly to the amp’s output terminals, and its gain set to maximum.
Now, consider how deep into the future we are, that this is now the kind of thing a person might actually do. You’re listening to an amplifier buzz, over the internet. 😀
I don’t mean to overstate the case, or even suggest that the GFA-535 is a particularly noisy amplifier. All amps produce some noise that can be heard if you put your ear near the speakers. But the 535 is noisier than other classic Adcoms, especially in the right channel. With efficient speakers, the buzz can be quite noticeable.
Warning: the following information is extremely boring if you are not an electronics technician. For most people, it’s enough to know that these improved, quieter power supply boards are now standard with my audiophile restoration packages for the GFA-535 and GFA-545, MK1 and MK2
Still with me?
It always seemed a bit odd to me how the power supply in the 535 is wired, especially the routing of the signal ground. Questions about it have come up on discussion forums, but I had not seen a satisfactory explanation.
People wonder why this amp module signal ground terminal is not connected on the right channel…
…but it is connected on the GFA-545! The amp boards are identical between these two models, but the 545 does not use a power supply board. It has a traditional power supply, with chassis-mount capacitors, and a star-ground with terminals for everything. This method works really well, actually.
But on the 535, the signal ground for the right channel is instead conducted through the shield of the coaxial wire leading to the RCA input jacks, where it is tied to the left channel’s ground, which does have it’s signal ground connected back to the power supply. So basically, the inputs are starred together, before they go back to the star through just the left channel’s side of the power supply. Hmm…
This diagram shows the OEM layout in simplified form. It does not quite show the subtleties of the star-grounding system. Mostly just ground paths are shown.
Point “E2L” on the bottom left of the power supply board is used as signal ground for both channels, conducted through the input coax shields. E2R is left open.
I can’t really know what the designer of this board had in mind, but it appears from the labeling of things, that the original intent was to use E2L and E2R to run to each E2 signal ground. This would buzz badly, as the transformer ground is connected right along the conductor between points E2L and E2R, so these would not be good points to reference signal ground. Maybe the idea was just to use only one E2 point at a time, but if this was the case, E2 is still not a great spot to take signal ground. It doesn’t have it’s own path back to the star, but shares a conductor with the right channel’s signal ground, currents flowing through the drain resistor, and tiny charge pulses from the 0.22uF poly bypass cap. (C905 and R901)
The original board:
Long story short, this board is not optimal. The way the amp is wired, appears to me, to be a factory production “Mod” to deal with a board error. I don’t mean to be too critical of 1986 Adcom; these are beautifully designed amplifiers. Mistakes happen, and every product ever made has bugs. It’s certainly no fault of the amp’s circuit designer, Nelson Pass. He provided the amplifier’s design, but would not have been asked to design the power supply board. Also, there are space limitations, single-sided board limitations, and economic limitations. (Doing it better takes more parts.) You can see how much board real-estate is gobbled up by those speaker connectors in back. I’m not sure I could design a good board with a single layer, in that little space.
My ideas about the actual causes are really only hypotheses at this point. I have not proven or isolated the actual mechanisms. If you see any errors in my logic, then I am very interested in hearing about it, and will revise the article accordingly. Please contact me with the form below. Live long and prosper.
The basic issue is that there are some long loops here to pick up hum and interference. For one, the signal ground for the right channel wends a long and winding road along the coaxial input cable. It’s serving double-duty as an audio conductor, as well as signal ground. It does at least effectively reference both channel’s signal inputs grounds to each other.
But E3 on each board is Zobel ground, and the Zobel RC network happens to be AC coupled to a point between the amplifier output and the feedback resistor. This is a point where noise can be injected into, and amplified by the negative feedback circuit.
The left channel’s Zobel ground is correctly referenced to a star point in the middle of the power supply board, and its signal ground is pretty much correctly starred to this same point. Relative to the signal ground star point, this is pretty OK. There’s not much potential difference between these two points. The feedback network is unperturbed.
The right channel is a different story. While it references signal ground from the same point as the left channel, E2L, it’s idea of zobel ground is quite different, and does not agree with its own signal ground. There is a potential difference between the E3, and signal ground. If you trace between E3 and signal ground, that is all the way through E3L on the PS board, through the star ground point in the middle of the board, which is polluted with rectifier currents, and on down to E2L, signal ground. This path is a noise generator, as well as an induction pickup. This noise is coupled by the Zobel to the negative feedback circuit where it is amplified. Because of the inverting nature of negative feedback, it manifests as odd-order harmonic noise.
Another issue is the way the grounds for both channels are simply tied together. There will necessarily be some potential difference between the two grounds. If tied together with a low impedance ground plane, some big currents can flow. The dual transformers makes it worse, as these will also have some potential difference between their center taps.
A common way to deal with all this is to insert some small resistance between each channel’s power supply ground to chassis ground, usually 10 to 100 ohms, in parallel with a small poly capacitor, 0.1 to 1uF, to conduct RF. This works great, but leaves the unit unprotected from power supply faults such as a short from transformer primary to secondary, or a power supply lead touching the chassis. In the event of a power supply fault that shorts to chassis, the resistor simply pops open. (They use flame-proof metal oxide resistors for this reason.) The chassis may go live if it is not earth-grounded. It may also blast current across the shields of the RCA cables to other equipment. In amps with this arrangement, I often find this resistor is blown, leaving the chassis unshielded.
A more modern, and safer solution is to additionally place a bi-directional diode-shunt across this resistor and capacitor. The shunt does nothing in normal operation. But if a fault occurs, the diodes will clamp down hard at about 1V. In a major fault, the diodes could feasibly blow out, but when they do, they usually short out, which is fine. The mains or power supply fuses should blow. For redundancy, one can double up the diodes, and a bridge rectifier just happens to have four robust diodes in a convenient package. The bigger these diodes, the better. They should be able to handle much larger surge currents than the fuses. I use a flat-pack style 35A.
Here’s how my new board is wired. I have not invented anything new. These are just best-practices, as far as I know, for wiring a dual-mono amplifier. It is essentially two mono amps in one chassis. Again, this is a simplified diagram and does not show the subtleties of star grounding.
Design of the star-grounding system on the PCB is surprisingly involved. One must consider an array of currents traveling through conductors, and the point at which you take the signal ground reference must be chosen carefully, to be as clean and still as possible. Watch where the all the currents flow and give them their own paths wherever possible. There are spiky rectifier currents charging the capacitors, speaker returns, pulses from bypass caps and drain resistors, etc… For this board, I did most of the star-grounding on the top layer, and I extract the precious signal ground from the bottom of the board, where it joins with the Zobel ground, away from all the cap-charging chaos above-board. This is basically like stacking ring terminals on a traditional star-ground ground bolt. Signal ground on top of the bolt, then zobel, then caps and rectifiers.
It’s made of double-thick 2oz copper for lower resistance. The two bridge rectifiers in the middle are the diode shunts, and the regular bridge rectifiers use the holes in the front. The original board had no heatsinking for the bridges, and I sometimes see these bridges blown out from overheating. In my design, they are mounted below the board, and bolted to the chassis floor with thermal grease. The 11mm hole allows access with a nut-driver. The bridge leads are soldered once the board is secured in place, and the whole assembly can still be easily removed later.
The two small blue caps in the front are X2-rated snubber capacitors for the AC side of the bridge.
You may notice I am taking advantage of space on the board originally occupied by connections for these lousy speaker terminals…
…which is fine. I always replace these with a single set of nice binding posts, and also remove the speaker switches. The holes these posts leave behind are kind of large, and normal binding posts won’t fit. You can either plug the eight holes with little plastic plugs, and mount the new binding posts elsewhere on the rear panel, or just use binding posts wide enough to cover the holes, with a little epoxy to keep them centered.
A word before we go on…
Please do not attempt any project inspired by this post without fully understanding how to safely wire an AC-powered appliance. If you are going to modify the AC wiring of your amplifier, please remove the two-wire power cord, and install a new 3-wire power cable or power inlet jack, with a safety-earth connected to the chassis. Above all other considerations, it is vital that there be a direct, uninterrupted and low-resistance connection from the mains safety earth ground wire, connecting directly to the amplifier chassis with a crimped ring terminal, and locking bolt. If this single connection is done correctly, then you or anyone who comes into contact with your equipment are much better protected from potentially fatal mistakes. The safety earth connection exists to ensure the chassis never becomes energized with a dangerous current.
The loop breaker I describe here must never be used between safety earth and the chassis. It is strictly for use from the amplifier’s power supply ground to the chassis. The earth ground lead from the power cord or input jack should be wired directly to a crimped ring terminal. (It can be soldered additionally but must be crimped too.) The ring terminal should be bolted to an un-painted spot on the chassis, and secured with toothed lock washers. There may be specific regulations for this connection where you live. This earth ground bolt should not be used for any other connection or purpose. It should not be shared with the mounting flange of another component like a transformer or capacitor clamp. Chassis earth ground is a connection a life may literally depend on, so make it a good one.
Anyways… so serious… results???
Before the new board, I could hear the buzz coming out of the speaker with my ear a few feet away. Now, with my ear right up to the speaker, I can barely hear anything at all! I had to check the amp was actually on. It’s an amazing improvement! Sweet, sweet silence. Now, let’s quantify the improvement.
Noise at the speaker output of an amplifier is in the sub-millivolt range, making it difficult to get meaningful measurements from a multimeter, scope, or much of anything but a very expensive spectrum analyzer. What’s needed is a low-noise amplifier to boost the signal enough to analyze with the computer.
I happened to have just the thing lying around. An M-Audio MicroTrack 24/96 portable digital recorder. It has a very quiet microphone preamplifier, a low noise floor, and records uncompressed 24/96 wave files. I made a nice shielded input cable with alligator clips to connect to amplifier terminals.
This is a pretty ancient device these days, but as you can see, the noise floor is extremely low and smooth, even at maximum gain. This is plenty of signal-to-noise ratio for analyzing an amplifier’s hum at the speaker terminals.
I’m using RightMark Audio Analyzer to do the FFT. It’s mainly designed to benchmark sound cards and the like. It’s free software, yet features a pretty amazing FFT analyzer! 8 million points, if you should ever want that many. That’s around the resolution of a top-of-the-line spectrum analyzer, though limited to the audio band. I used 1 million points, Hanning window, for the FFTs.
So, the moment of truth: I made a recording of my own GFA-535’s speaker outputs, with the inputs shorted. First, wired as original. Then, I installed the new board and took a new set of measurements. Here’s before and after, with the “after” shown in green, and shifted to the right. It’s a little hard to read because of the overlay.
The left channel. A nice improvement! The fundamental and second harmonic are down a couple db, but the higher harmonics are down as much as 8db! Woo!
But that’s nothing. Check out the right channel!
Notice how it’s the odd-order harmonics that are massively reduced. Most of the even-order harmonics are actually slightly increased, even though the overall noise level is much lower. The large odd-order harmonics seen with the OEM arrangement supports the hypothesis that the noise is getting injected into the negative feedback loop.
Amazing. Hooked up to speakers, there is near silence. No buzz. I have to put my ear right up next to the speaker to hear anything at all.
Besides the FFT, just listening to it shows the obvious improvement. The levels in the recording are unchanged.
It’s not just less buzz, it’s actually more pleasant to listen to, with its even-order harmonics.
I also have board for the GFA-535 MKII, and GFA-545 MKI and MKII. These boards now come standard with every GFA-535 or 545 audiophile refurbishing package. The 535II models will not show as dramatic an improvement in noise. Measurements on those boards coming soon.
Boards are available for $50. Installation is an intermediate level electronics project with some danger of shock from mains voltage. Please be careful. Read and watch all the internet things on the subject. Rod Elliot covers the topic thoroughly as usual.
Contact me here if you want one… I’ll provide some more detailed installation instructions in the future, but in the meantime, we can discuss the finer points via email or phone.
I’m buzzing! This is a great improvement to an already legendary amplifier. Thanks for reading.
New board version!
I did some experimentation with my own GFA-535, moving the ground points around on the circuit board, and measuring the change in noise. I discovered that it’s best to tie the zobel and signal ground together before connecting to the power side of the star. I created a new version, and they arrived today!
Even better! Even quieter! Check out the before and after FFTs.
The previous version showed some spikes around 11K, and these are now gone almost completely. Could it have been something like a ringing in the zobel network? With the new board version, the zobel and signal ground are agreeing more closely because they are tied together first.
If you’d like to order a board, contact me here. They are $50 each, for any of the three versions. (535, 535II, 545I or II)
I will offer a complete parts kit in the near future, but for now I can supply a recommended parts list.