Technics SL-PG390 Mods
The Technics SL-PG390 is almost identical to the more expensive SL-PG490, which includes a headphone amplifier with volume control, and comes with a remote control. The 390 will work with the same remote control if this is bought separately. The SL-PG390, 490 and 590 were introduced 14 years ago, and there should still be plenty available in good working order, but there appear to be more PG390s around, so I decided to use this model. The SL-PG590 at the top of this range has a different case which includes anti-vibration damping, but uses mostly the same components. The PG390 case contains mostly empty space, which makes additions and modifications easier.
A good place to start is with a copy of the Technics service manual. A free download can be found at: manualslib.
The signal processor is listed in the manual as a MN662713RG1, but I could only find data for the similar MN662710RA and MN66271RA.
At least one company, Atoll, thought the Technics model was worth 'improving', and the review of their CD 100 at tnt-audio suggests that this £1000 player is little more than a standard PG490 (photos here), with a different output stage instead of the original BA4560 op-amp (Similar to LM833). The description says this new stage is 'dual-mono', 'symmetrical' and has a 'low impedance' output (750 ohms, compared to 1k for the Technics). Also it is supplied from a 165VA torroidal transformer. The output stage that is, the original transformer still remains for the rest. Why the output stage needs such a big transformer is not explained, and I can see no heatsinks to suggest that high power levels are being dissipated anywhere. The original player has a specified total power consumption of only 13W. One possible way to reduce vibration is to add mass, so maybe that big transformer does something useful after all. As for 'symmetrical' circuits, this I suspect is just a marketing ploy, if it helps at all symmetry usually only reduces even order harmonic distortion, not the more unpleasant odd harmonics. Maybe they just mean the layout is symmetrical.
First of all I wanted to determine whether the original player starts out with any obvious measurable problems. A test disc is the first requirement. The following is a source of suitable files:
The Audio Check page includes a two-tone intermodulation test signal generator for which any frequencies and levels can be chosen, but the duration is limited to 10sec. The result can then be downloaded as a wav file for inclusion on a test cd. Using 'Nero Express' or something similar the wav files can be burnt as an 'audio cd'.
Nero has an option to check the result for errors compared to the original files, but I have never had any errors reported. Even the cheap dvd burner in my computer evidently has no problem both burning and reading discs without errors. It is not clear whether only uncorrectable errors are reported, but either way it appears reading the data from a disc can be extremely reliable, unless the disc has significant damage or manufacturing faults. The different error correction used for audio discs and data discs should be irrelevant, the Nero error check still has to read the audio disc correctly to carry out its test.
We can probably discount most of those 'tweaks' related to the actual reading of the audio data from the disc. Uncorrectable read errors certainly can occur, but these are not so common that reducing them is going to transform the sound quality. I saw an estimate somewhere of one uncorrectable error per 190 hours of music being achievable, and also data collected by AccurateRip which shows that for the better computer drives there are as few as one music track out of 156 with any errors. Those tweaks certainly have no direct link to jitter, the data is read and then stored in a buffer memory and read out at a rate determined by a crystal oscillator. My SL-CT700 portable player has a much bigger buffer than most normal players so that for example it can be used when out jogging, and any rapid movement which interferes with data reading has no audible effect. Even violently shaking the player while playing a CD has no obvious effect. At the best quality setting its buffer length for a conventional music CD is 10sec. Such extreme measures are not needed for a 'static' player, and the buffer length is probably much lower, the PG390 processor data just specifies the existence of 'anti-shock functions'.
I had planned to compare the PG390 to a 16-bit Marantz player, I said somewhere that I prefer the sound of my Panasonic portable player to the Marantz, and the PG390 seems to have the same advantage, whatever it is, real or imaginary. I may be half deaf, but I don't have much imagination either, and so I had expected to find a real problem with the Marantz, maybe it just has a boosted treble response, or maybe the DAC linearity has deteriorated over the years. Unfortunately the Marantz refused to play my test discs, which may point to some more general disc reading problem. I may try a different brand of CD-R sometime, but for now here are a few results just playing a test cd through the PG390. First, to check the frequency response, this is 1kHz plus 20kHz both at -12dB. For digital signals it is convenient to specify peak levels relative to the 'full scale' level. The -30dB test signal level in the picture is just a convenient level for the EMU1820m soundcard. A range of frequencies were tried, confirming that the response is adequately flat.
The next tests reveal some low levels of distortion. First a 1kHz sinewave at -2dB, followed by 1kHz at -60dB. For this second result the soundcard gain is turned to maximum.
The highest distortion component in the -60dB spectrum is 3rd harmonic about 47dB below the 1kHz test signal, and so 107dB below the full scale level. These components could be from the D/A converter, or from the analogue output stage if it is suffering from significant crossover distortion, or a combination of both. The distortion specifiation for the output op-amp seems to be missing from the data sheet, but for the similar LM833 is 0.002% at 3V output and increasing at lower levels, which may just be noise or maybe crossover problems. I checked this track on the computer drive via the EMU card and confirmed that the distortion is not already present on the recording. There was only a very low component at the 3rd harmonic visible.
I have found a -60dB 1kHz spectrum published some years ago for the Marantz CD-63, which I believe is a better model than my old CD-273SE. This can be found on page 15, Fig3 of an article (pdf file) about differences between 'numerically-identical' CDs. The 'distortion' components are around the same level as my results for the PG390, but the big difference is that the Marantz components are mostly not at harmonics of the 1kHz test signal. The text suggests that these effects are caused by interference from the servo and digital circuits. It is encouraging that the Technics result is mostly just harmonic distortion.
That article also suggests a few other tests including a check for a type of 'jitter' which can be caused by a particularly difficult test signal. I tried this, but the results were inconclusive, there seemed to be missing high order harmonics rather than the sort of added non-harmonic components shown for the Marantz player, and the result is shown next followed by the near perfect result from the computer drive plus EMU soundcard. Note that the PG390 output includes components below the -120dB level, demonstrating that the often quoted 96dB dynamic range limit of audio CDs is not really a limit. This is however not an entirely normal music signal. I have added a footnote with further explanation.
Next is an intermodulation test with 19kHz plus 20kHz both at -12dB. The 1kHz intermodulation product is at -100dB relative to the 20kHz component.
Next an oscilloscope trace showing a 1kHz 'square wave'. This doesn't look very square, but this is about what we should expect if the square wave has all components up to 21kHz included with correct amplitude and phase, and all higher frequency components completely removed. It is sometimes suggested that a result with flat tops and rounded corners is in some way better, but to achieve that I believe it is unavoidable to add phase nonlinearity in the audio band. The only benefit would be if subsequent stages were unable to handle the peak levels, which are slightly higher with the 'ringing', but this should never be a problem in any competent design.
The PG590 has better THD specifications than the PG390, reduced from 0.007% to 0.004%. Looking at the circuit diagrams, this would make a great 'spot the difference' competition. They are almost identical, but having opened up the 390 I see that the very helpful manufacturer has used the same board for all three models and marked the locations of unoccupied holes with part numbers and notes about which models those parts are used in. That suggests a good starting point for an upgrade.
Looking first at the left hand photo, D861 is part of the muting circuit, which is not needed. Adding C402 (470u 6V3) and Q401 (2SD1862QRTV6), and removing link J123, provides separate supply smoothing for different circuits, so may help reduce interaction between the audio and other functions. There is another small difference in the circuits, which is R413, used in the 390 but not the 590. It is not obvious either what this does or why the models differ, but it is evidently not needed when modified. The polarity of the electrolytics must be correct, the thick curved white lines on the board indicate the negative terminals.
The right hand photo shows the location of C20 and C21 (100u 25V) which add supply smoothing for the audio output stage op-amp. If a complete separate output board is added this is not needed, but the simpler option of using a better op-amp may benefit from adding these capacitors. There are two missing links also visible in the photo, J310 and J308 which are just an earth connection for the headphone amplifier and so don't need to be added. There are other differences comparing the 590 and 390 but most are because of the additional headphone amplifier and more complex muting, so not necessary for the upgrade.
It is tempting to follow the Atoll path and add an elaborate discrete component output stage, but just trying a simple op-amp change may demonstrate whether this is solving any real problem. The D/A converter has its own internal op-amp at its audio output, and so if this or the DAC itself is causing the distortion observed there is little to be achieved by replacing an innocent op-amp.
I have been looking for jitter figures for the Technics players, so far without success, but the specification in the manual of 'wow and flutter' as below the measurable limit is encouraging at least. One option is to add supply smoothing at pin 60 of the DAC which is the crystal oscillator supply. This could in principle reduce jitter, but there is already extensive supply smoothing at various points, and some measurements may be a good idea before risking damage to the fine tracks around the 80 pin LSI. How to measure jitter may be the problem to solve first before trying some uncertain and possibly unhelpful addition.
The plan is to try some of these ideas and see if there is any measurable difference. Some 'tweakers' base their modifications on their listening experiences, which may or may not be reliable, but as always my own approach is to rely more on theory and measurement. I have no idea how tweaks which make no measurable improvement can be judged, there is little chance of any such 'improvements' passing a properly conducted double-blind listening test, so it then just ends up as one person's opinion, which is fine for that person but not necessarily much help to anyone else. How can I choose an op-amp based on listening opinions when almost any reasonably good op-amp has its fans who will go on about how much the soundstage expanded or how they heard instruments they never noticed before, or how product X wipes the floor with product Y. Anyway, I am already happy enough with the sound quality, this is just an interesting project to see how far a low cost player can be improved. Have corners been cut to keep the cost down, or is it already about as good as it can get without a drastic redesign?
The first result is for the added supply smoothing already used in the PG590, requiring the addition of one 2SD1862TV2Q (Farnell part no.1680149), two 100uF 25V electrolytics (1219466, Panasonic FM) and one 680uF 16V (1848385, Panasonic FM). The 680u should really be a 470u 6.3V, but I forgot to order that. Increasing the values is always tempting, but unexpected consequences such as switch-on thumps are then a danger, so it is safer not to stray too far from the Technics values. Link J123 and R413 are removed.
I had planned to do a few measurements after just these changes, but having the board out and having a LM4562MA op-amp I decided to add this also. Big mistake! Trying to remove the old op-amp it seemed very hard to move, melting the solder down one side and levering it up with a screwdriver it wouldn't lift off the board. Eventually I realised it was glued to the board. Having wrecked a track or two and probably the op-amp too with excess heat, I had to break it away from the board. Anyone trying this removal should perhaps try a different approach, maybe desoldering each pin in turn and twisting it up from the board before finally using force. I also added a small spot of glue to hold the new op-amp in place before soldering it in place. Two tracks were bad enough to need wire links adding. Putting everything back together I was astonished to find it still worked, and even sounded alright. Anyway, that reminded me why I hate surface-mount components. The first of the next photos shows how not to do it.
Here is the result at 1kHz -60dB after these changes.
Comparing this to the earlier result it is disappointing, the harmonic distortion is about the same. The op-amp already has its own supply regulation, so adding two capacitors was not likely to help much, and having used the LM833 in the past I know it is quite good, so the similar BA4560 op-amp was unlikely to be a problem. The splitting of the 5V supply is less easily dismissed as an improvement, it may be meant to benefit the oscillator or control circuits, so may reduce jitter even though distortion is not noticeably different. Why the PG590 has a better distortion specification remains unexplained, but there is an inevitable suspicion that this is another marketing ploy to persuade buyers to choose the more expensive model. An alternative less cynical possibility is that the DAC boards were tested for distortion and the best samples used in the top model.
I have replaced the output muting transistors with a relay, Farnell part number 9913920. At first I found that although the voltage across the coil was switching on and off at the right times the relay was not operating. I thought it must be a faulty relay, but looked more closely at the data sheet on the Farnell website and found the terminal assignment diagram, which shows positive polarity on pin 1 and negative on pin 10. I had the reverse polarity, having assumed it to be unimportant, but for this particular relay it matters, and reversing the connections it started working as intended, the click when it operates is clearly audible. The next diagrams are before and after modification. Confusingly the relay top view is shown, but as seen in a later photo I mounted the relay with its pins upwards.
Looking at the circuit diagram on page 17 of the manual, Q851 can drive the relay, which has a coil resistance of 1800 ohms and is nominally 12V but needs a minimum of 9V for operation. The existing load, R852 is 2k2 so removing that there is little difference in operating current. The '10.4V' supply however changes under some conditions, and may fall close to 9V, so for added security I took one end of the coil to the +14.5V op-amp positive supply, at the positive terminal of the recently added C20. The other end of the coil goes to the collector (the middle pin) of Q851. Q853 must be removed, together with the muting transistors Q801 and Q802. Everything else can be left alone. To prevent damage to the switching transistor from back-emf a diode should be connected across the relay coil in such a direction that it is reverse biased when the coil is activated, i.e. with the cathode (usually marked with a band) at the terminal connected to the positive supply. I used a 1N4001. The contacts are connected so that with the relay off the tracks which originally connected to the muting transistor emitters are connected to earth, which can conveniently be the original transistor collector points. The signal is therefore muted until the relay is activated, so there is no delay during switch-on before the muting takes effect.
If the original muting transistors were causing any distortion we would expect the greatest effect at high signal levels, so I checked the distortion at high levels after fitting the relay, but the results looked almost identical to those with the muting transistors, so again any benefit is not apparent.
Having concluded that most of the 'improvements' appear not to be doing anything useful, I decided on one, for now, final and not entirely serious example, following from the Atoll upgrade. I just happened to have a faulty torroidal transformer I was going to put in the bin, so instead it is now bolted to the case through a conveniently placed hole. The player is now far heavier, and if vibration was ever a problem it could now be significantly better. Using a torroidal transformer to power an elaborate discrete component output stage however seems somewhat superfluous when the original op-amp appears to have been almost entirely harmless, practically all the low level distortion being produced somewhere earlier in the signal chain.
The next photos show the relay and the final (for now) version with the mass-adding transformer.
My conclusion is simply that none of the easy and obvious modifications are worth doing, there is no clear benefit from any of them. If I listen long enough maybe I can convince myself that I hear an improved soundstage, or can hear sounds I never noticed before, but beyond some possibly psychological benefit I think I have to say the player was already pretty good, and short of completely replacing the digital processor with something comparable to that in the EMU soundcard, or using the optical output with an external DAC, it already was about as good as it can get, and I will give up for now and just play some music. I think at least I have done no noticeable harm, it still sounds good to me.
Audio CDs use 16 bit data, which gives a theoretical range of 96dB, so some may be puzzled by the frequency spectra showing a much higher signal range. Here is an extreme example showing the spectrum of a signal played from an audio CD on my computer drive, with a peak at 0dB, a series of signal components around -130dB and a background noise level apparently close to -180dB, so where has that extra 80dB range come from?
An averaging function has been used to reduce the noise, but even so this is an astonishing result. The lines are all spectra of square waves, one at 11.025kHz, exactly a quarter of the 44.1kHz sampling frequency, the other is at 229.6875Hz, which is the sampling frequency divided by 192. The difference compared to a similar result shown earlier is that I disabled the analogue input of the soundcard, so that the input amplifier noise is avoided. This result has nothing to do with dither, often used to reduce quantisation distortion, as far as I know no dither was used for this signal. To see the explanation think of a low level low frequency square wave where the level changes by just the least significant bit (LSB) of the digital signal. The fundamental frequency will be around the -96dB limit of 16-bit audio, but it is still a square wave and its harmonics still exist at odd harmonics reducing in level as we go up through the frequency scale, so for example the 49th harmonic is at 11.255kHz and is at a 49th of the fundamental level, so about 1/49, or 34dB below the supposedly -96dB minimum level, i.e. at -130dB, almost exactly where we can see a component just after the 11.025kHz 0dB signal.