The original 1960’s PA234 IC’s (and the Sylvania/NTE rebadged versions) are long since obsolete and are now almost impossible to find. This chip is utilized as an output channel amplifier (and small 25 Ohm speaker driver), as well as a spring reverb driver in these instruments. Because of this scarcity, both the mk1 VCS3 and Synthi A are seriously compromised should any of these chips fail in them. This is a very good reason to try and make a discrete version available as part of their future-proofing.
Note that the PA234 is a silicon monolithic IC (not germanium, as some have speculated) and was part of a series of silicon power audio (‘PA’) amplifier IC’s that General Electric made in the late 1960’s, including PA222, PA234, PA237 and PA246:
https://www.semiconductormuseum.com/Transistors/GE/OralHistories/Jones/Jones_Page4.htm .
Many people have commented on the ‘sweeter sound’ that the mk1 VCS3 and Synthi A appear to have compared to the later mk2 Synthi’s. The mk2 dropped the use of PA234 entirely and utlized a 741/complimentary germanium transistor push-pull hypbrid amplifer instead. Whatever the perceived sweetness of the sound is, it is not due to germanium, but maybe precisely because it is all-silicon! Of course one should not forget the different circuit topologies used in each case, which might also be a factor.
Some advantages of a discrete version:
- No special parts. The design makes no use of any matched transistors etc and uses generic parts.
- It’s more stable to overvoltage compared to the PA234 IC. The PA234 has a maximum rail to rail voltage of 25v. Anything much higher can destroy them. This can happen in the mk1 Synthi, if the BFY51 transistor providing +11.3v on the positive rail of the 3 PA234s fails with the collector-emitter shorted, which then puts +19v on the positive rail on them, which, along with the –9v negative rail, gives a differential voltage of 28v. This can lead to their destruction.
- By careful design, the discrete version can be built on a small pcb with DIL pins which will allow it to neatly fit into a 14pin DIL socket on board A of the mk1 Synthi, using the existing PA234 solder pads. This also allows for easy exchange of PA234 or the discrete version, which is useful in repairs. If an original PA234 cannot be immediately found, the discrete version can, at the very least, be used as a stop gap.
The main disadvantage (at least for the v1 prototype built and tested thus far) is that it consumes more current compared to the PA234. However, the total current draw of 3 v1 discrete devices is still well within the capabilities of a heatsinked BFY51 transistor that provides the +11.3v positive power supply for the output channel and reverb drivers in the mk1 Synthi.
In the following discussion, ‘PA234’ will denote the original monolithic PA234 IC, and the discrete version will be denoted as ‘DPA234’.
The original PA234 1W amplifier on a silicon chip
This is a redrawing of the PA234 schematic, taken from its datasheet:
The schematic is quite typical of a quasi-complimentary class A-B amplifier. The driver stage is formed by npn transistors Q1 and Q2. The output stage by the npn Darlington Q4 and a Sziklai ‘pair’ consisting of the pnp driver Q3 and another npn Darlington Q4.
Such amplifiers were common in the days before decent discrete pnp silicon transistors could be made cheaply. As mentioned, PA234 is a silicon monolithic device. Manufacturing pnp transistors on a silicon wafer was not cheap in the 1960’s. This probably explains why the device uses 6 npn and only 1 pnp transistor in the quasi-complimentary topology, as shown.
The 3 diodes in series create an ‘auto-biasing’ of the output stage transistor bases such that even in quiescent mode both are biased to be active. This is a rudimentary way of avoiding the usual crossover distortion that push-pull amplifiers suffer from, without having to setup more complex (but better) biasing networks. It keeps parts count minimal which probably was desirable in designing an audio amp on a chip for the (non-audiophile) consumer market!
Resistor R1 limits the current through the biasing diodes and the collectors of Q1 and Q2. Resistors R2 and R3 at the output are included to minimize distortion.
The schematic (like the PA234 chip) requires an external biasing/feedback resistor network to set the biasing of the driver stage and the DC level of the output along with capacitor coupling on the input and output to protect this biasing. An example may be found in the datasheet, but note this is NOT the same biasing circuit as used in the mk1 Synthi.
The advantage of this amplifier design (there is no differential pair) is it does not need on any special matching of transistors nor concern about their precise gain values when designing a discrete version. The actual ac gain of the amplifier is dependent on the choice of external biasing resistors (as explained in the datasheet).
The Discrete PA234: DPA234
A 3d render of the populated v1 pcb is shown above. To keep the overall size down to something not much larger than a 14 pin dip IC (like the PA234) some smd parts are used.
Testing the DPA234 v1
The real test of the DP234 is how it sounds compared to the original PA234 when used in the mk1 Synthi. This was done by using the E.M.S Synthi KB1 which has a mk1 board A and uses PA234 for the 2 output channel amplifiers and the spring reverb driver.
All 3 PA234s were removed and 14 pin dil sockets soldered to board A (which in any case is a good idea for their easy replacement, if ever needed).
The method of testing involves choosing a single output channel and to play various patches and waveforms from the KB1 and record the sounds and analyse the resulting scope waveforms and recordings. For each patch or waveform, this was first done with DPA234 inserted and then immediately after, with the PA234 inserted. Using the same output channel avoids any slight channel-to-channel variations in e.g., the tone circuits or FET level control between output ch1 and output ch2.
Fixed Waveform Testing
The first test used sine and square wave fixed frequency waveforms generated by the KB1 (just the raw wave forms, no filtering). The scope grabs below show the sine wave output through DPA234 and PA234. The purple plot above each is a FFT which gives an indication of the harmonic content. Apart from very small differences in amplitude the two output waveforms are a close match.
The plots below shows how the sine wave is clipped when the level of output ch1 is pushed to its maximum. Apart from a few mV differences in amplitude, the two clipped waveforms are very similar.
The plots below are scope captures for a square wave from the KB1 which again shows very close matching.
Sound Files
In this test, some short recordings of different patches on the KB1 were made, alternating between the DPA234 and PA234 being in place in output ch1 amplifier, so a direct comparison can be made. Audacity software was used to make the recordings, running on a Macbook Air. The test recordings are listed below.
Frequency respsonse DPA234 vs PA234
Even without analysing the sound files, but just listening, there is a close match between the DPA234 and PA234 output amplifiers.
To confirm this in a more quantitative way, a swept sine wave was taken from a signal generator and fed into input ch 1 of the KB1 and from there directly to output ch1 (via the tone circuit, which was set at midpoint). The sweep range was 100Hz through 10kHz over a 10sec period at 600mv p-p amplitude.
To compare the frequency response of the two sweeps, the ‘Plot Spectrum’ analysis tool included with Audacity was used. The results are shown below.
The relative frequency response of the two amplifiers are shown below.
Current draw and Thermal Stability
The Darlington transistors Q4 and Q5 run hot to the touch even in the quiescent state, with no input signal. It’s important to check that their case temperatures do not approach their absolute maximum ratings. Their maximum junction operating temperature is 200oC which is higher than most TO-92 package transistors. While the case temperature is typically less than the junction temperature, on the small E-line packaged transistors used, the difference is likely not that much, so the case temperature is a good guide of the junction temperature.
The following positive supply current draws were measured, along with the Darlington transistor case temperatures (using a digital thermometer with K-type thermocouple probe):
Quiescent: Current draw through positive (11.3v) supply rail: 33mA
Case temperature: 50oC
At maximum output level: Current draw 50ma
Case temperature: 60oC
Even a case temperature of 60oC allows the Darlington to operate well within its maximum rating.
BFY51 transistor: current load and case temperature
On the Synthi mk1 board A, a single BFY51 transistor provides the +11.33v positive rail voltage for all 3 PA234’s. It does this via a simple emitter follower circuit which takes +12v and diode drops it to +11.33v at the emitter. The collector voltage is around 19v.
Thus, with all 3 DPA234’s fitted we have at a total quiescent current draw of 3x33ma = 99ma. The quiescent power dissipation of BFY51 is therefore (19-11.33) x 0.099 W = 0.76W.
At maximum sound levels through the KB1speakers, the total current drain through BFY51 increases to around 150ma and the total power dissipation is {(19-11.33) x 0.15 }W = 0.92W. Case temperature measurements of the BFY51 were:
Quiescent: 61oC
Maximum level of Output channels: 63oC
The data sheet of BFY51 states a maximum power dissipation of 2.5W if the case temperature (via a fitted heatsink) is kept below 100oC. So, the conclusion is that the total current drain of 3x DPA234 discrete amplifiers still allows for both the Darlington transistors and the BFY51 in the +11.33v positive supply to run well within their maximum values of power dissipation and junction temperatures.
Future work
Whilst the v1 DPA234 works very well and reaches thermal stability without excess current draw, it would be nice to try and run it cooler. This would certainly reduce the current load (by how much remains to be seen). To this end I am investigating an all-smd version of the DPA234 with copper pour areas connected to the Darlington transistors heatsink tab.
I have built the all-smd version prototype and this adesign allows to add a small heatsink bonded with thermal glue to the pcb. The heasink works well and gets warm to the touch. Apart from dissipating heat from the output transistors, it also helps keep the diodes in thermal equilibrium with them. This is important as the diodes are crucial to avoid potential thermal runaway (by providing negative feedback). For this to work efficiently it’s better they are close as possible to the temperature of the output transistors.
The all-smd pcb is a little longer than the previous version, but they do still fit the original Synthi mk1 board A:
I have tested the new version with max level signals being pushed through both output channels and the reverb continuously for several hours. The heatsink reaches thermal equilibrium even in this extreme case.
Constantin at portabellabz will also conduct further testing (my thanks to him for discussions during this project and also for his prevous extensive testing of the v1 pcb). It looks like this will hopefully be the final version of the discrete amplifier.