More Power Amplifier Surgery
Repairing MOSFET Power Amplifiers
Power Amp Surgery
Although the operation of MOSFETs is radically different to Bipolar devices, the repair procedure is broadly along the same lines. Thankfully, repairs are much less onerous, as chain destruction rarely occurs, whilst testing the devices is, if anything, easier than testing Bi-polar transistors, particularly as basic checks can be done without removing the output devices. Moreover, locating replacement FETs is easy, as there's a large degree of standardisation in device parameters. Now in order to avoid unnecessary iteration of the finer details of common dismantling and test procedures, it's assumed that you've a copy of E&MM June '82 to hand as a cross-reference.
The straightforward circuit, shown in Figure 1, applies (give or take a few refinements) to a large number of MOSFET power amplifiers from UK and Japanese manufacturers. This simplicity has a large bearing on the reliability and audible quality of the best MOSFET designs. Amps which don't share the common topography of Figure 1, tend to be based on classic (i.e. Complex!) Bipolar designs with 20 or more small signal transistors. Here, the Bipolar output devices have been exchanged for MOSFETs, and a few other modifications will have been made, notably (a) to suit the new biasing and DC level shifting requirements, and (b) to compensate for the difference in phase/stability margins when slow Bipolar output transistors are exchanged for much faster FETs. In this latter species, repair techniques are a mixture of those presented here and those appertaining to Bipolar Amps (see E&MM June '82). We'll therefore concentrate on Amplifiers having a circuit topography similar to Figure 1.
Firstly, in the vast majority of cases, catastrophic failure will be limited to the MOSFETs, and even here, failure is usually limited to only one or two devices. To test the MOSFETs, you'll need a 'voltage source' ohmeter. Some early digital multimeters don't present a significant voltage across their probes when switched to 'ohms', but more recent types have a 'diode test' facility which rectifies (!) this shortcoming. In fact, testing is easiest with two ohmeters, but only one of these will need to source a small voltage, circa 1½ to 2 Volts. In contrast to Bipolar amps, testing can begin without removing the devices from the PCB. Simply unhook the gate resistors from the remainder of the circuitry (RG in Figure 1).
For P channel FETs (the ones between the negative rail and output) the procedure is as follows:
1) Connect the positive lead of the 'Voltage Source' Ohmeter (VSO) to the gate and the negative lead to the source. Note here that we're assuming real polarity - on a typical analogue meter, the positive lead on the ohms setting will be the one that's normally negative for voltage and current measurements. This procedure discharges the gate, and a near infinite resistance should be noted.
2) Reverse the leads: Negative meter lead goes to gate and positive to source. This charges the FET, and again, the resistance should be near infinity. If not, the MOSFET is faulty.
3) Assuming an infinite resistance is recorded, leave the VSO in position (i.e. keep the FET charged) and connect a second meter across the source and drain terminals. A reading of 1 to 5 ohms indicates all is well; if the resistance is significantly higher, then (a) the VSO meter isn't sourcing sufficient voltage to turn the FETs hard on or (b) the FET is faulty. If you only possess one meter, charge the gate and then move the negative lead from gate to drain. This must be done quite smartly, because the gate's charge will progressively leak away through the PCB and even the surrounding air if the atmosphere's humid enough.
4) Then discharge the gate (+ve lead to gate, -ve lead to source), and check that the drain-source resistance rises to a high value - above 10k at least.
5) If the board has several paralleled FETs on each side, you'll have to charge and discharge all the devices on one side (e.g. All the p-channel devices) before making the first drain-source measurements. If there are then any discrepancies in the drain-source readings, you'll need to remove all the devices or at least cut the source or drain tracks in order to track down the faulty device(s).
For the n-channel devices, repeat the process, simply reversing the polarity of all connections.
Figure 2 displays this procedure in easily assimilated visual terms for use at the test bench.
Having located faulty MOSFETs, the next task is to secure suitable replacements. Unlike Bipolar transistors, the determination of equivalents and sourcing are both relatively simple. The key manufacturer of audio grade MOSFETs is Hitachi, and as implied by the contents of Table 1, it's only necessary to determine the polarity, package and voltage (VDSS) rating. The latter parameter is determined in essentially the same manner as the VCEO for a Bipolar output device, viz: 10% to 30% higher than the total supply rail voltage. For example, devices with VDSS = 160 volts are ideal for amplifiers having +70volt rails. A proviso is in order here, however. In amps using a single MOSFET on each side, the relatively high 'on' resistance (cf. a Bipolar power device) circa 1 ohm spells a drop of several volts across the device, and so the other FET doesn't have to sustain the total rail voltage.
As a result, in a 'one FET-per-side' amplifier, some manufacturers have got away with devices having a VDSS equal to or fractionally below the total rail voltage. This sort of skimping can also be found in amps with paralleled FETs; in this case, it works usually until the day after the guarantee runs out!) because the Hitachi ratings are conservative, and most devices breakdown at a voltage somewhat higher than the quoted VDSS. Clearly, this isn't a good recipe for reliable amplification, and if the original devices have a 140V VDSS with +70 volt supply rails, this probably accounts for the amplifier being on the repair bench! If in doubt, always aims for generous VDSS ratings. MOSFETs with VDSS in excess of 160V are presently rather expensive, so until their price falls, it may be expedient to reduce the supply rail voltages by adjusting the transformer tappings if the rails are too high for 160 Volt devices.
Before replacing the faulty FETs, be sure to thoroughly clean away any residue, soot and general gunge left in the wake of the recent expiry! Replace the gate resistors, apply fresh thermal mounting compound and, when each device is mounted, check for isolation between the case and heatsink with an ohmeter before moving on to the next replacement. Also lookout for slivers of wire nestling in the thermal compound. Once the faulty MOSFETs have been replaced, remove the drive and current source transistors (TR3, 4 and 5 in Figure 1) and test these. If they're intact, it's fair to assume that TR1/2 are also OK. If not, check all the transistors, diodes and resistors and replace any which are suspect. In addition, replace the gate protection zeners (where fitted - they're really superfluous in most cases) as a matter of course. Next, carefully check the PCB tracks for damage, solder bridges and odd whiskers of wire lying between tracks. Finally, turn the quiescent current preset (PR1) for minimum resistance between the N and P channel gates.
Once you're satisfied that everything's in order, reconnect the amplifier to its power supply via 470R, 7 watt resistors (in series with the positive and negative supply leads), plus a milliammeter in series with the positive lead. In addition, monitor the supply rail voltage across the PCB terminals. After making final checks, turn on. If the resistors heat up and the voltmeter reading fails to rise above 10 to 20 volts, this suggests that PR1 has been turned in the wrong direction, or a fault, so switch off and double check. If, on the other hand, all's well, switch off and replace the series resistors with something smaller (say 100R, 7 watt) and switch on again. Check that the quiescent current (IQ) can be turned up to quite a high value (500mA) by adjusting PR1. Don't be surprised however if this level of IQ causes the voltage across the board to sink to zero! (100R x 500mA = 50 Volts) providing IQ can be wound back to a few milliamps only, and in this case, nearly the full rail voltage is indicated on the voltmeter, then it's hunky dory. Switch off again, remove the series resistors and reconnect the power. This time, set IQ (by adjusting PR1) to 60-100mA (1 FET per side), 120-200mA (2 FETs per side) or 250-500mA (4 FETs per side). The exact value isn't particularly critical, but it's wise to steer to the lean side unless the heatsinking is exemplary and the amplifier boasts fan cooling. If a scope is available, and the heatsinking is good, adjust the quiescent current to remove any signs of secondary crossover distortion at high frequencies (20kHz-30 kHz) whilst the amp drives at half power into a low impedance test load. Remove the milliammeter before making this test if you ever aim to use it again!
Now turn the gain control to minimum (or short the in put terminals) and look at the DC voltage across the output. If this is below 500mV or so, all's well; over 1 volt suggests that something's amiss, and accordingly, switch off and check the board, in particular C1, which may be leaky or reverse polarised. If in doubt, try (a) a replacement or (b) a reversal. On some boards, there may be a DC offset-nulling preset hung around the emitters of TR1/2 or thereabouts. If so, you can tweak this for a near zero offset.
The next test requires a scope. MOSFETs tend to be prone to instability in the 1-10MHz region (Cf. 30-200kHz for Bipolar amplifiers), and whilst the oscillation is usually quite small - 4 volts peak-to-peak or so - and essentially inaudible it can make the top end sound wrong, and make low level listening fatiguing its presence is obviously unsatisfactory. Assuming the devices you use as replacements are pretty close to the original (which is likely), then oscillation isn't inevitable, but if it does occur, it can nearly always be cured by tweaking the values of CC (usually in the downwards direction) and CS by +100%. If the Bipolar driver transistors have been replaced, it's also worth tweaking CO. Note here that the reference of the heatsink and the quiescent current setting both govern stability. Therefore, if the heatsink is normally grounded to chassis (say), and you're testing the amp's board outside of the chassis, be sure to link the heatsink to the chassis with a short, stout jump lead before jumping to conclusions about the alleged oscillatory stroppiness of MOSFETs! Equally, be sure to tweak the quiescent current either side of the setting you've chosen just to be sure that the newly 'cured' oscillation isn't still lurking...
On the topic of instability, here's some remedial surgery that applies equally to Bipolar transistor amplifiers: If the output choke, RL or the Zobel network components (CZ and RZ) have burnt out or show signs of having become very hot, then either RF (Radio Frequency) noise has been fed up the amp, or severe instability has arisen somewhere in the system. RF burnout is a frequent cause of demise in Rock 'n' Roll amplifiers, especially the 'fast' varieties with extended top end response. It's very unlikely that you'll make such an amplifier go unstable on the bench, but as soon as you put it on stage and connect up all the cables, lo and behold, acrid smoke ariseth... Counter measures against this antisocial habit are 3 fold. Firstly, increase the value of CR to attenuate the amplifier's sensitivity to RF appearing at the input. This capacitor forms a -6dB/Octave filter in conjunction with RH, and the -3dB point should be set no lower than 25kHz (or so) for decent top end performance. Of course, if the amp in question is always used on bass and/or midrange only, you can lower the -3dB breakpoint considerably. This frequency (FO) can be calculated approximately as follows: FO = 1/6 x CR x RH, where C is in uF and R in ohms. For example, the network shown in Figure 1 has a breakpoint of 76kHz.
The second, palliative measure, is to increase the capabilities of the output network. RL and RZ should be carbon resistors. Wirewound types are used by some manufacturers because that's an easy way to achieve high power ratings. However, even 10 watt wirewound components, with instantaneous ratings of 100 watts or more, are found to shatter or rupture, and moreover, these components fail at RF because of their inductance. Instead, use 2 watt carbon resistors. This is the highest wattage commonly available in carbon film. Double their value and parallel up to give 4 watts. For example, RZ could be uprated with two 22R, 2 watt resistors. Strangely, this 4 watt carbon arrangement is far less prone to self-destruct that a high power wirewound component, but if in doubt, you can parallel up quite a few 2 watt units, providing you keep the leads as short as possible. In a similar vein, C2 should be exchanged for a high voltage polypropylene unit: this species of dielectric is capable of withstanding large RF voltages without overheating and consequent fire risk - it's not unknown for polyester capacitors in this position to explode in a sheet of flame!
A more positive action, but one involving lots of investigation, empirical tweaking and sundry black magic is to locate and cure the source of RF. If it's generated within the amplifier, but not on the test bench, careful examination with a scope in the field and some adjustment of the output network values will often secure a solution. If all else fails, the sledgehammer - a 'secondary zobel' - as arrowed in Figure 1 mounted across the output socket usually succeeds - try 100n and 4R7 as initial values. If that fails, you'll have to settle for a Gremlin exorcism...
Last of all comes the soak test. Apply pink noise (or a signal from an off station FM tuner) and drive the amplifier into a dummy load at quarter power (half the maximum voltage swing) for several hours. If the amplifier has fan cooling, more rigorous testing can take place at half power (0.7 x the maximum voltage swing), which will stress the thermal performances of the output devices to the maximum. With this gruelling test, it's a good idea to monitor the heatsink temperature, for unless the heat dissipation qualities of the amplifier are very good, the 'burn in' exercise can easily turn into a spectacular burnout.
The procedures in this article refer to 'S' type MOSFETs. Over the next year or so, 'D' type MOSFETs will begin to appear. These types have a much lower RON, and required a higher voltage - some 9 volts or so - to turn them on and off. As the average analogue meter, or digital meter with diode test only supplies 1.5 to 4.5 volts on the ohms range, a modified test jig will be required, consisting of a PP3 battery, a series current limiting resistor (say 1k) and a milliammeter.
Some of the new generation FETs (whether 'S' or 'D' types will appear in a plastic 'T03' package and the leadouts may be reversed relative to Table 1. FETs with higher voltage and power ratings will also become cheaper and more readily available over the next 12 months and will enable amps under repair to be uprated at the same time.
Despite this volatile situation, help is at hand to E&MM readers. Comprehensive information, advice and actual replacement MOSFETs can be supplied by a Specialist High-Tech dealer: Pantechnic, (Contact Details).