There is a lot of information on these forums about the merits (or not) of replacing the standard opamps in DACs & CD players with something "better". Unfortunately not many threads focus on the technical aspects and some more in-depth analysis of discrete vs. IC opamps. I decided I had to try for myself and contribute my findings, for what it is worth.
I could lay my hands on a few DEXA discrete opamps at a discount, and I am in the lucky possession of two Marantz CD80 CD players, one of which I bought used recently and which was in need of major servicing. Hence this was a good opportunity to experiment with some discrete opamps. This way I could use one of the players as a reference with the original NE5534's left in place.
This article will focus on both the “rejuvenation” of the Marantz CD80 CD player, as well as the experience of replacing the standard opamps with either OPA627 IC opamps or DEXA discrete opamps.
Subject Equipment
Test Equipment
Listening & Reference Equipment
I don’t think it is a good idea to roll opamps without having access to equipment to check for both stability and bias levels. No doubt many would disagree, but IMHO an oscilloscope (ideally dual channel, 50MHz or above) is required when rolling opamps, even when the manufacturer claims that they are "drop-in" replacements. Typically the stock opamps are replaced with "something faster", which could lead to stability problems, potentially leaving you with an AM or FM transmitter. This would not necessarily be audible on the audio output.
For CD players I would also suggest to use a test disk with some test tones at different frequencies and levels, as instability may not necessarily occur unless the opamps are driven at normal operating levels, which can manifests itself typically as a "fat" trace on an oscilloscope.
The Opamps
One of the opamps used is a NewClassD discrete opamp (DEXA single), which the manufacturer claims to be a drop-in replacement for a range of industry-standard opamps thereby significantly improving the sound quality. We shall see. Although NewClassD made their name designing Class D amplifiers, rest assured that the opamps are not class D.
Notably absent is the availability of a data sheet. Few data are given on these opamps other than that they are a novel 12-transitor design featuring an extremely linear and fast response, therefore allowing for a low gain. I am not going to argue with the designer of these marvels, but most circuits employing opamps are designed around the typical high gain they provide.
Permissible voltage range (+/- 5V to +/- 25V), typical supply current (7mA) and the dimensions (width 13mm x height 28mm) are all the data available on these opamps. Other than that, we are unfortunately pretty much left with the manufacturer's statement that these are drop-in replacements for a whole range of listed opamps.
The opamps themselves are fortunately fairly small as far as discrete ones go, and in reality measure in at 11mm width, 24mm long and 30mm tall (older versions are a bit taller). Some shuffling of components near the opamp may still be required. On the CD80 for example the decoupling caps had to be moved a couple of millimetres. The opamp has no external compensation or balance facility, and is unity gain stable.
The other (integrated) opamp used is the Burr Brown OPA627, a well-known Difet input opamp with a considerably higher bandwidth than the NE5534. Fortunately a datasheet for this is readily available, as it should be. The opamp is unity gain stable (the OPA637 is its brother intended for use with noise gains greater than 5). There is no external compensation (pin 8 is not connected), but it has a balance facility. Note that the typical supply voltage is 15V with a maximum limit of 18V, so you can't just drop them in anywhere to replace a 5534 which can tolerate up to 20V. The CD80 provides 16V, so we're good.
The Marantz CD80 player(s)
The players used are Marantz CD80s from the late 80's. In fact the samples I have are from 1989 and 1990 respectively, with some very minor revisions made to the main PCB in between. This was Ken Ishiwata's first CD player design for Marantz, so I have been told. At the time it was a competent implementation of the classic Philips SAA7220P/B 4x oversampling filter & TDA1541A (S1 single crown) DAC combination.
The die-cast chassis, power supply and transport are based on the Philips CD880, the rest of the player is Marantz' own design. It employs separate regulated analogue & digital power supplies (discrete for the analogue section) and the main PCB has separate digital & analogue ground planes (joined at the filter-DAC interface). The regulators are far away from the receiving components, but that does have the advantage that the considerable heat generated by this power supply can be dissipated through an externally mounted heatsink. Furthermore the power supplies have reasonably generous distributed decoupling (Elna Cerafines) close to the receiving components which makes it not quite as bad as it would look at first sight.
It has a solid die-cast metal Philips CDM1-MkII swing-arm mechanism with Hall motor. These almost never die, and if they do play up it is rarely a problem with the laser. Usually the electrolytic caps on the servo board need replacing, or the CD hold-down bearing needs polishing / reconditioning. If the laser does need replacing, spare CDM1-MkII mechanisms will be almost impossible to find, but as long as the Hall motor spinning the CD is still in good order the swing arm containing the laser can be transplanted from pretty much any CDM4 mechanism, which are easily obtained used. In short, this thing can in theory be kept going almost forever.
Finally, an additional 8 kg or so of base-metal thrown in for the chassis, bottom & cheeks bring the total weight to about 15kg. Hence over-engineered, but not necessarily in all the right places.
Preparation; Rejuvenating The CD Players
Both players were in original factory condition, both obtained used. The younger of the two players was literally completely disassembled to remove all the smoke & nicotine deposits (my major gripe with used equipment, nicotine pervades everything!).
First all 23 year-old electrolytic capacitors in both players were replaced with new ones. Elna Cerafines were replaced with identical new Elna Cerafines, and all other generic Philips and Nichicons replaced by Nichicon KZ, ES and FG where appropriate. Only the two main Elna Cerafine 6800 microFarad, 35V supply caps could not be obtained, they are OEM only. These were replaced by Nichicon KG Gold Tunes. Not quite as good as Cerafines, but new Gold Tunes have got to be better than Cerafines that have been baking near a heat sink for 23 years.
The laser current was adjusted, as on both players it had drifted about 20% too high (monitor at 60mV whereas it should be 50mV +/- 5mV as per spec; the monitor on most Philips CDM servos is dimensioned to give about 1mV reading per 1mA laser current).
Both the focus lenses and fixed lenses were cleaned, with care these can both be accessed on most Philips CDM swing arm mechanisms by removing the focus lens armature. Even if the lenses look clear to the eye, they may have deposits that inhibit the invisible infrared light. A cotton tip with distilled water and naphthalene are my preferred cleaning agents. Isopropyl alcohol would probably be safe too although this might affect the lens adhesive or coatings used.
The CD hold down bearing surfaces were re-polished and lubricated. Some Philips CDM mechanisms of this era have a hold down clamp that can wear a small indentation into the flat plastic bearing surface pressing down on it. This can make the whole drive shake if the CD hold down is not exactly centred relative to the motor axle (it has a little play). Particularly at the first track of a cd this seems to coincide with the drive's suspension resonance frequency, which has a tendency to throw the tracking off course (skipping). Removing the indentation solves this. Difficult with the type of plastic used, but I have had good success using a diamond file, followed by 1200 grit sandpaper and finished with paper polishing.
The oversampling filter was left as is, as the only way to really improve the relatively high jitter in these players is to remove the SAA7220 altogether, fitting a new clock. This would also disable the oversampling and additional sample error correction offered by this filter, and I am not a big fan of losing that. I only mention this to qualify that my impressions on the impact of using the DEXA and OPA627 opamps are in context of jitter levels that would be bettered by most modern players. I do not think the jitter is something I could identify in a blind test, but I am not denying there are people with better ears out there who could.
Finally both players were given a few days of running in of the new caps, and to my ears both players sounded pretty similar after this (and rather a lot better than with the original caps!). In a blind test I would not be able to tell both CD80s apart. These reconditioned CD80's are in fact rather nice CD players, and won't be leaving my abode until broken beyond repair!
Rolling The Opamps
The CD80 employs four JRC NE5534D opamps, two per channel. The first two, immediately following the DAC current source outputs, take care of the I/V conversion and any required de-emphasis filtering. The second two act as unity gain buffer amplifiers ensuring a stable output into varying loads and also shield the DAC from any “mishaps” that might occur on the user end of the RCA sockets. An overall DC feedback servo removes any residual DC component on the output. It does so by biasing the DAC current source output voltage (nominally 0mV) feeding into the first opamp.
The four NE5534s were carefully removed and turned IC sockets put back in place. The 22pF compensation caps allowing the NE5534s to work at unity gain were left in. The DEXA opamps do not use these (nor does the OPA627) and leave the corresponding compensation pins floating. The DEXAs are designed to be unity gain stable (they have to be as a simple pair of the DEXA opamps in a dual package are available for replacing NE5532s).
The DEXAs do have almost double the typical current draw of the NE5534s, and logic dictates that the ripple current drawn by them could also be higher. Therefore the Elna Cerafine 100 microFarad decoupling caps around each opamp were replaced by Elna Cerafine 220 microFarad ones, which have a higher ripple current rating (probably overkill, but can't do any harm).
The power feed resistors were left unaltered (33 Ohm 0.3W) as these are fusible safety resistors. The capacitor substitution therefore moves the corner frequency of the local opamp power supply filter from 50Hz to 25Hz, which also won't do any harm. The DEXAs themselves simply push into the fitted IC sockets.
First a check with an oscilloscope & CD with test tones (don't feed this to your speakers or headphones!), and all seems stable for all frequencies and signal levels. I was using a 70MHz oscilloscope, so anything up to a couple of hundred or so MHz instability would be observable (albeit at attenuated levels). This was tested using both an open output and typical RCA cable as loads, I won't be using anything else in practice.
First Listening Session
I am pretty open-minded, but I would like the result to sound better. Problem is: it doesn't. It is hard to put my finger on it, but something is amiss. Something that a hundred hours of running-in time isn't going to fix. The detail is all there, and the highs do sound a little less "grainy", but I cannot deny that the unmodified CD80 with four NE5534s has a certain coherence to the sound that is lacking in the modified player utilising four DEXAs. Imaging is also poorer than in the stock version. PRAT (Pace, Rhythm & Timing) are, frankly, shot to pieces.
Hard to describe, but if I were pushed I would say it now sounds like a first generation bitstream player. Oh dear. What could be wrong?
More Measurements & The TDA1541A Compliance
Out come the oscilloscope and multi-meter again, and it doesn't take long to spot at least one potential problem. IC opamps have a well matched differential input stage, and combined with a high open-loop gain generally only require a very small difference in the inverting and non-inverting inputs to affect a significant change in output voltage.
Not so the discrete DEXA it seems. It evidently has relatively poorly matched input transistors. I say relatively as this may be as good as it gets in a discrete design. The DAC's current output has been pulled up to +30mV. That doesn't sound like a lot, but he TDA1541A's output is a current source with a compliance of only +/-25mV maximum.
In practice I know this is ignored by a lot of modders (or they are not aware?), but Philips clearly states that in order to have the TDA1541A meet its stated performance levels, maximum permissible compliance is +/-25mV. This is a mid-80's Philips assertion and if audiophiles have learned anything in the intermediate years it is that to meet our current audiophile standards often we need to apply more stringent criteria on top of official component specs. Hence I would advise to assume the TDA1541A was designed to dump its current into a very low impedance indeed.
DEXA Gain-Bandwidth Product And Differential Input Matching
Not being quite content with the lack of data available on the DEXAs, and despite the limitations of my measuring tools (3MHz function generator & 8-bit digital oscilloscope with max. 2mV/Div sensitivity), I decide to have a go at measuring the gain-bandwidth product of the DEXAs.
Setting up a non-inverting unity gain configuration (with 22nF compensation cap), I measure the gain-bandwidth product of the original JCR NE5534D at a little shy of 20MHz. This is higher than the typical spec of 10MHz, but probably within production tolerance. The DEXA measures in at around 8MHz (based on a single sample I should add, I don't have a large batch of DEXAs lying around to work out an average!).
The NE5534 follows the 20MHz gain-bandwidth product fairly accurately up to 100kHz, but somewhere between 100kHz and 1MHz it falls off a cliff. At 1Mhz the gain-bandwidth product is not much more than 1MHz, or a gain of one. I suspect this has a lot to do with the 22pF compensation capacitor employed (as per the CD80's configuration). I also suspect that if I were to set up the NE5534 for 10dB gain or so (avoiding the need for the compensation capacitor), it would manage to maintain a good gain at 1MHz.
Although the DEXA has a lower gain-bandwidth product of around 8MHz, it does manage to maintain this comfortably up the 3MHz limit of my function generator. It is notable however that whereas the NE5534 setup clearly struggles at higher bandwidths, the DC imbalance between the inverting and non-inverting inputs is stable at 0mV. The DEXA however shows a high, frequency dependent drift in this imbalance, up to about 20mV or so. This again is bad news with respect to the TDA1541's 25mV compliance.
Differential Input Stage Matching And Temperature Drift
As would be suspected, further measurements show that the imbalance between the inverting and non-inverting inputs of the discrete DEXA is not only far greater than that of an IC opamp, but also very sensitive to temperature variations. In fact, although the DEXAs have a DC offset trimpot (tiny, and sensitive!), it is proving very hard to get the DC level of the DAC current output reliably close to zero Volt, especially when playing some music.
I know from experience that the CD80 internals run quite warm, and bringing up the internals to the operating temperature makes the DC offset on the DAC's current output drift by a good 15 to 20mV. Effectively this is a result of both the I/V feedback topology around the first DEXA with poorly matched differential inputs as well as the DC feedback servo doing its best killing the DC offset following the second DEXA. There are two DEXAs used more or less "in series" and unfortunately it appears the temperature sensitivities of both DEXAs in this particular configuration are additive rather than cancelling.
Combine that with the very sensitive trimpots (also a sign of future drift & stability problems), the frequency dependence of the imbalance, plus a 3mV or so ripple on the DAC's current output when playing a 20kHz 0dB signal level, and it soon becomes evident that keeping the DAC output reliably within 25mV compliance is going to be hard.
The DEXA Topology
These revelations have made me rather curious about what this “novel” DEXA circuitry actually looks like. Again, a schematic of the DEXA internals is not available, unlike for many other opamps (at least a search on the web for the circuit diagram was unsuccessful). It looks like we are going to have to do it the hard way!
The DEXA PCB is very small with SMD components. However, with a magnifier and a good light source it is possible to spot the traces on both sides of the PCB. Quite a few hours later I finally have a drawing of the PCB layout, and some time later in front of me lies a complete drawing of the DEXA circuit. This was not an easy exercise! Out of respect for NewClassD’s IP and as I don’t accidentally want to fall foul of any copyright legislation I will not reproduce the circuit here.
However, what I will reveal is that the design features a degenerated complimentary differential input stage employing four transistors. As this would ideally require the use of two matched pairs of transistors, this may help explain the relatively high imbalance and temperature sensitivity between the inverting and non-inverting inputs.
Further Considerations
The observations so far suggest that these DEXAs aren't quite the "drop-in" replacements they are claimed to be, at least not for a CD80. My guess is that this would be an issue with most TDA1541A based DACs, especially when employing a DC feedback servo. Furthermore, with the poor matching of the differential inputs of the I/V DEXA, the de-emphasis filter (part of the I/V stage) may also miss its target correction filter characteristic (I can't easily test that though).
Finally, the CD80's DC servo contains a couple of polarized electrolytic capacitors just outside the DAC output, feeding to ground via a resistor. As per the original design these are happy having virtually 0mV permanently biased either way when using the stock 5534s. However, with the DEXAs this could easily be up to 30mV reverse bias or so, depending on the operating temperature. In theory the caps should survive that, but I would nevertheless feel a little nervous having polarized electrolytic caps permanently reverse biased to that amount. Therefore, if the DEXAs are left in the I/V stage I would certainly replace these caps with bi-polar versions.
Rolling Only The Buffer Opamps
So, now what? Obviously one option is to keep the 5534 opamps in the first I/V stage, and use the DEXAs only in the second buffer opamp stage. Two DEXAs pulled and TI NE5534AP I/V opamps back in. O-scope tests show this configuration to be unconditionally stable as before, but the DAC outputs are now at a reassuring stable 0mV (well at least less than 1mV), even when playing a 0dB signal or any music. Heating the internals up makes no measurable difference whatsoever. Good.
Adjusting the trim pots on the remaining two DEXA buffer opamps to get the output of the preceding 5534 I/V opamps close to 0mV (feedback circuits are funny aren't they), and we are ready to go. This trim isn't really necessary as the 5534s are quite happy running with a DC offset of a few mV on their outputs, and the signal between the 5534s and DEXAs has no (practical) compliance limitations. It is just a bit of "tidying up" really.
Much Better
OK, this is better. Much better. The "coherent" sound, for lack of a better way of describing it, and PRAT are back. But (and this is even better news) the somewhat "grainy" treble of the stock CD80 has given way to an equally detailed, but more refined sound. The imaging is now also a tad better than with the unmodified CD80.
After a few days running-in time, things have only improved further. The sound is more transparent, has better imaging and more depth. The stock CD80 does have a subtly "warmer" sound though, but I would rather suggest the DEXA modified one is more neutral, especially when compared to the SA11S3, which it now comes a little closer to than does the stock CD80. The only "issue" is that it has made the Sennheiser HD800's even more unforgiving with respect to poor recordings. Good recordings however sound better than ever. As I said, the Marantz CD80 has not yet become a Marantz SA11S3, but it is getting closer now.
Rolling The I/V Opamps For BB OPA627
The next thing to try is to use a well-matched IC opamp alternative for the NE5534 in the I/V stage. Despite their relatively high cost, I have seen the Burr Brown OPA627 recommended time and time again. It is unity-gain stable, and the CD80 supply voltage is just about right, so we give it a go.
The OPA627BP opamps were selected and ordered from Farnell as they have slightly better characteristics than the OPA627AP, but are virtually the same price. Both of these get you to about half the price level of a discrete opamp, so they are still more than ten times the price of the humble NE5534. As these opamps have FET inputs they do require ESD handling precautions.
Swapping them in place of the NE5534 is otherwise straightforward. These opamps are significantly faster than the NE5534, so again operational stability was verified using a test disc and oscilloscope as before, which did not reveal any problems. The current output of the DAC was running at an expected negligible offset (<1mV). Incidentally, the 22pF compensation capacitors were left in as, just like the DEXA, the OPA627 does not have compensation pin 8 connected.
Much Better Still!
The OPA627 has a reputation for sounding a little laid back, if anything. I have to agree. In fact I would say the overall balance is now pretty close to the stock NE5534 / NE5534 combination. Typical somewhat laid back but nevertheless refined and engaging Marantz sound. So, was this then a very expensive way to get back to from where we started?
Fortunately, with this OPA627 / DEXA combination there are important aspects of the sound that have improved significantly. The CD80 now sounds really, really nice indeed. Whereas the stock CD80 at times can sound like it is struggling a little, this combination sounds effortless. A well balanced, neutral yet engaging sound with great PRAT and not a hint of listening fatigue. Interestingly, with this combination the Sennheiser HD800 is now again a little more forgiving of less-than-ideal recordings.
Although this is still not up to the SA11S3 level, it does manage to capture a good slice of the SA11S3 "magic" compared to the stock CD80. Whereas with the stock CD80 you always hear that you are listening to a very good CD player with a pleasing sound, both the modified CD80 and SA11S3 at times manage to make you feel the musicians are standing in front of you, or that you are actually in the church with the choir. Philips' 1989 recording of Jose Carreras in Ramirez' Misa Criolla is a perfect example. The modified CD80 manages to resolve higher frequencies more cleanly, and some of the harshness of the stock CD80 is gone. The acoustics of the recording environment are better defined, church acoustics in Ton Koopman’s Bach Organ works for example sound more life-like.
In fact it seems that particularly low-level sounds like the background sound and reverberations in a church or castle hall sound significantly more life-like, this being achieved without the need for emphasis on treble (the frequency response is still flat).
Furthermore, on the stock CD80 voices (particularly "s"-sounds) can sound a little "smeared" as far as positioning is concerned, whereas both the modified CD80 and SA11S3 are able to pinpoint them more stable and accurately.
The SA11S3 does give the impression of a somewhat wider soundstage, and generally still sounds a little more refined than the modified CD80. However, considering that the SA11S3 is still between five to ten times the cost of a used CD80 + cap & opamp modifications, the latter is a very attractive package if you are after a good example of the SAA7220P/B & TDA1541A sound. In that context, starting off with a factory configuration CD80, the £50 (~$80) capacitor rejuvenation is definitely worth it, as well as the £100 (~$160) opamp upgrade if you can afford it. And a Marantz CD94MkII could well be an even better basis, if you don't mind modding a legend...
Conclusions
When rejuvenating an older CD player (or other audio component), the priority should definitely be on replacing the dried-out old electrolytic capacitors. Where they are in the signal path (and that is not always clear-cut!) you can put some better quality ones in, or at least match what was in there already. When it comes to the decision on whether to replace the critical opamps with either IC or discrete ones, I have found that this choice may well depend on whether the circuit, and especially the DAC, can tolerate the poorer input stage matching of discrete opamps. Off course you can always combine IC and discrete ones, and the OPA627 & DEXA combination tried here works very well indeed for the CD80 in this case study. Whether the result justifies the financial outlay of these more expensive opamps very likely depends on the particular audio component. If the design is poor to start with it would likely be a waste of money, but in case of the Marantz CD80 it is definitely a worthwhile investment.
UPDATE 25-Feb-2014:
The other CD80 now has had the same opamp treatment, with the same improvements resulting.
The next step will probably be to replace the headphone amp in one of them (currently a pair of NE5532s). There is plenty of space and I have a spare pot & headphone jack salvaged from a broken CD60, which are identical to the ones used in the CD80. I will probably design & etch a couple of boards for just the amp section of something like a Lehmann linear, or improved version of my own DIY effort from 1992. I just need to check the CD80's power supply has enough spare capacity for the slightly elevated currents required.
I could lay my hands on a few DEXA discrete opamps at a discount, and I am in the lucky possession of two Marantz CD80 CD players, one of which I bought used recently and which was in need of major servicing. Hence this was a good opportunity to experiment with some discrete opamps. This way I could use one of the players as a reference with the original NE5534's left in place.
This article will focus on both the “rejuvenation” of the Marantz CD80 CD player, as well as the experience of replacing the standard opamps with either OPA627 IC opamps or DEXA discrete opamps.
Subject Equipment
- A 1990 build Marantz CD80 for opamp rolling (fully recapped Nov 2013)
- Four NewClassD DEXA discrete single opamps
- Four Burr Brown OPA627BP opamps
- Four TI NE5534AP (2013) replacements for swapping back should the roll be unsuccessful
Test Equipment
- A dual channel 70MHz digital oscilloscope
- A high impedance digital multimeter
- A 3MHz Function generator (for opamp measurements outside the CD player)
- A CD with test tones at various frequencies and signal levels
- Marantz CD80 service manual
Listening & Reference Equipment
- Another 1989 build Marantz CD80 with stock JRC NE5534D (1989) opamps for reference (fully recapped Oct 2013)
- A 2013 build Marantz SA11S3 SACD player
- A 2013 build Marantz PM11S3 Amplifier
- A pair of 2013 build Spendor S3/5R2 monitor speakers
- A 2013 build Lehmann Linear headphone amplifier
- A DIY design & build discrete headphone amplifier (1992 build, fully recapped Oct 2013)
- A Sennheiser HD800
- A Sennheiser HD650
- A Sennheiser HD540
I don’t think it is a good idea to roll opamps without having access to equipment to check for both stability and bias levels. No doubt many would disagree, but IMHO an oscilloscope (ideally dual channel, 50MHz or above) is required when rolling opamps, even when the manufacturer claims that they are "drop-in" replacements. Typically the stock opamps are replaced with "something faster", which could lead to stability problems, potentially leaving you with an AM or FM transmitter. This would not necessarily be audible on the audio output.
For CD players I would also suggest to use a test disk with some test tones at different frequencies and levels, as instability may not necessarily occur unless the opamps are driven at normal operating levels, which can manifests itself typically as a "fat" trace on an oscilloscope.
The Opamps
One of the opamps used is a NewClassD discrete opamp (DEXA single), which the manufacturer claims to be a drop-in replacement for a range of industry-standard opamps thereby significantly improving the sound quality. We shall see. Although NewClassD made their name designing Class D amplifiers, rest assured that the opamps are not class D.
Notably absent is the availability of a data sheet. Few data are given on these opamps other than that they are a novel 12-transitor design featuring an extremely linear and fast response, therefore allowing for a low gain. I am not going to argue with the designer of these marvels, but most circuits employing opamps are designed around the typical high gain they provide.
Permissible voltage range (+/- 5V to +/- 25V), typical supply current (7mA) and the dimensions (width 13mm x height 28mm) are all the data available on these opamps. Other than that, we are unfortunately pretty much left with the manufacturer's statement that these are drop-in replacements for a whole range of listed opamps.
The opamps themselves are fortunately fairly small as far as discrete ones go, and in reality measure in at 11mm width, 24mm long and 30mm tall (older versions are a bit taller). Some shuffling of components near the opamp may still be required. On the CD80 for example the decoupling caps had to be moved a couple of millimetres. The opamp has no external compensation or balance facility, and is unity gain stable.
The other (integrated) opamp used is the Burr Brown OPA627, a well-known Difet input opamp with a considerably higher bandwidth than the NE5534. Fortunately a datasheet for this is readily available, as it should be. The opamp is unity gain stable (the OPA637 is its brother intended for use with noise gains greater than 5). There is no external compensation (pin 8 is not connected), but it has a balance facility. Note that the typical supply voltage is 15V with a maximum limit of 18V, so you can't just drop them in anywhere to replace a 5534 which can tolerate up to 20V. The CD80 provides 16V, so we're good.
The Marantz CD80 player(s)
The players used are Marantz CD80s from the late 80's. In fact the samples I have are from 1989 and 1990 respectively, with some very minor revisions made to the main PCB in between. This was Ken Ishiwata's first CD player design for Marantz, so I have been told. At the time it was a competent implementation of the classic Philips SAA7220P/B 4x oversampling filter & TDA1541A (S1 single crown) DAC combination.
The die-cast chassis, power supply and transport are based on the Philips CD880, the rest of the player is Marantz' own design. It employs separate regulated analogue & digital power supplies (discrete for the analogue section) and the main PCB has separate digital & analogue ground planes (joined at the filter-DAC interface). The regulators are far away from the receiving components, but that does have the advantage that the considerable heat generated by this power supply can be dissipated through an externally mounted heatsink. Furthermore the power supplies have reasonably generous distributed decoupling (Elna Cerafines) close to the receiving components which makes it not quite as bad as it would look at first sight.
It has a solid die-cast metal Philips CDM1-MkII swing-arm mechanism with Hall motor. These almost never die, and if they do play up it is rarely a problem with the laser. Usually the electrolytic caps on the servo board need replacing, or the CD hold-down bearing needs polishing / reconditioning. If the laser does need replacing, spare CDM1-MkII mechanisms will be almost impossible to find, but as long as the Hall motor spinning the CD is still in good order the swing arm containing the laser can be transplanted from pretty much any CDM4 mechanism, which are easily obtained used. In short, this thing can in theory be kept going almost forever.
Finally, an additional 8 kg or so of base-metal thrown in for the chassis, bottom & cheeks bring the total weight to about 15kg. Hence over-engineered, but not necessarily in all the right places.
Preparation; Rejuvenating The CD Players
Both players were in original factory condition, both obtained used. The younger of the two players was literally completely disassembled to remove all the smoke & nicotine deposits (my major gripe with used equipment, nicotine pervades everything!).
First all 23 year-old electrolytic capacitors in both players were replaced with new ones. Elna Cerafines were replaced with identical new Elna Cerafines, and all other generic Philips and Nichicons replaced by Nichicon KZ, ES and FG where appropriate. Only the two main Elna Cerafine 6800 microFarad, 35V supply caps could not be obtained, they are OEM only. These were replaced by Nichicon KG Gold Tunes. Not quite as good as Cerafines, but new Gold Tunes have got to be better than Cerafines that have been baking near a heat sink for 23 years.
The laser current was adjusted, as on both players it had drifted about 20% too high (monitor at 60mV whereas it should be 50mV +/- 5mV as per spec; the monitor on most Philips CDM servos is dimensioned to give about 1mV reading per 1mA laser current).
Both the focus lenses and fixed lenses were cleaned, with care these can both be accessed on most Philips CDM swing arm mechanisms by removing the focus lens armature. Even if the lenses look clear to the eye, they may have deposits that inhibit the invisible infrared light. A cotton tip with distilled water and naphthalene are my preferred cleaning agents. Isopropyl alcohol would probably be safe too although this might affect the lens adhesive or coatings used.
The CD hold down bearing surfaces were re-polished and lubricated. Some Philips CDM mechanisms of this era have a hold down clamp that can wear a small indentation into the flat plastic bearing surface pressing down on it. This can make the whole drive shake if the CD hold down is not exactly centred relative to the motor axle (it has a little play). Particularly at the first track of a cd this seems to coincide with the drive's suspension resonance frequency, which has a tendency to throw the tracking off course (skipping). Removing the indentation solves this. Difficult with the type of plastic used, but I have had good success using a diamond file, followed by 1200 grit sandpaper and finished with paper polishing.
The oversampling filter was left as is, as the only way to really improve the relatively high jitter in these players is to remove the SAA7220 altogether, fitting a new clock. This would also disable the oversampling and additional sample error correction offered by this filter, and I am not a big fan of losing that. I only mention this to qualify that my impressions on the impact of using the DEXA and OPA627 opamps are in context of jitter levels that would be bettered by most modern players. I do not think the jitter is something I could identify in a blind test, but I am not denying there are people with better ears out there who could.
Finally both players were given a few days of running in of the new caps, and to my ears both players sounded pretty similar after this (and rather a lot better than with the original caps!). In a blind test I would not be able to tell both CD80s apart. These reconditioned CD80's are in fact rather nice CD players, and won't be leaving my abode until broken beyond repair!
Rolling The Opamps
The CD80 employs four JRC NE5534D opamps, two per channel. The first two, immediately following the DAC current source outputs, take care of the I/V conversion and any required de-emphasis filtering. The second two act as unity gain buffer amplifiers ensuring a stable output into varying loads and also shield the DAC from any “mishaps” that might occur on the user end of the RCA sockets. An overall DC feedback servo removes any residual DC component on the output. It does so by biasing the DAC current source output voltage (nominally 0mV) feeding into the first opamp.
The four NE5534s were carefully removed and turned IC sockets put back in place. The 22pF compensation caps allowing the NE5534s to work at unity gain were left in. The DEXA opamps do not use these (nor does the OPA627) and leave the corresponding compensation pins floating. The DEXAs are designed to be unity gain stable (they have to be as a simple pair of the DEXA opamps in a dual package are available for replacing NE5532s).
The DEXAs do have almost double the typical current draw of the NE5534s, and logic dictates that the ripple current drawn by them could also be higher. Therefore the Elna Cerafine 100 microFarad decoupling caps around each opamp were replaced by Elna Cerafine 220 microFarad ones, which have a higher ripple current rating (probably overkill, but can't do any harm).
The power feed resistors were left unaltered (33 Ohm 0.3W) as these are fusible safety resistors. The capacitor substitution therefore moves the corner frequency of the local opamp power supply filter from 50Hz to 25Hz, which also won't do any harm. The DEXAs themselves simply push into the fitted IC sockets.
First a check with an oscilloscope & CD with test tones (don't feed this to your speakers or headphones!), and all seems stable for all frequencies and signal levels. I was using a 70MHz oscilloscope, so anything up to a couple of hundred or so MHz instability would be observable (albeit at attenuated levels). This was tested using both an open output and typical RCA cable as loads, I won't be using anything else in practice.
First Listening Session
I am pretty open-minded, but I would like the result to sound better. Problem is: it doesn't. It is hard to put my finger on it, but something is amiss. Something that a hundred hours of running-in time isn't going to fix. The detail is all there, and the highs do sound a little less "grainy", but I cannot deny that the unmodified CD80 with four NE5534s has a certain coherence to the sound that is lacking in the modified player utilising four DEXAs. Imaging is also poorer than in the stock version. PRAT (Pace, Rhythm & Timing) are, frankly, shot to pieces.
Hard to describe, but if I were pushed I would say it now sounds like a first generation bitstream player. Oh dear. What could be wrong?
More Measurements & The TDA1541A Compliance
Out come the oscilloscope and multi-meter again, and it doesn't take long to spot at least one potential problem. IC opamps have a well matched differential input stage, and combined with a high open-loop gain generally only require a very small difference in the inverting and non-inverting inputs to affect a significant change in output voltage.
Not so the discrete DEXA it seems. It evidently has relatively poorly matched input transistors. I say relatively as this may be as good as it gets in a discrete design. The DAC's current output has been pulled up to +30mV. That doesn't sound like a lot, but he TDA1541A's output is a current source with a compliance of only +/-25mV maximum.
In practice I know this is ignored by a lot of modders (or they are not aware?), but Philips clearly states that in order to have the TDA1541A meet its stated performance levels, maximum permissible compliance is +/-25mV. This is a mid-80's Philips assertion and if audiophiles have learned anything in the intermediate years it is that to meet our current audiophile standards often we need to apply more stringent criteria on top of official component specs. Hence I would advise to assume the TDA1541A was designed to dump its current into a very low impedance indeed.
DEXA Gain-Bandwidth Product And Differential Input Matching
Not being quite content with the lack of data available on the DEXAs, and despite the limitations of my measuring tools (3MHz function generator & 8-bit digital oscilloscope with max. 2mV/Div sensitivity), I decide to have a go at measuring the gain-bandwidth product of the DEXAs.
Setting up a non-inverting unity gain configuration (with 22nF compensation cap), I measure the gain-bandwidth product of the original JCR NE5534D at a little shy of 20MHz. This is higher than the typical spec of 10MHz, but probably within production tolerance. The DEXA measures in at around 8MHz (based on a single sample I should add, I don't have a large batch of DEXAs lying around to work out an average!).
The NE5534 follows the 20MHz gain-bandwidth product fairly accurately up to 100kHz, but somewhere between 100kHz and 1MHz it falls off a cliff. At 1Mhz the gain-bandwidth product is not much more than 1MHz, or a gain of one. I suspect this has a lot to do with the 22pF compensation capacitor employed (as per the CD80's configuration). I also suspect that if I were to set up the NE5534 for 10dB gain or so (avoiding the need for the compensation capacitor), it would manage to maintain a good gain at 1MHz.
Although the DEXA has a lower gain-bandwidth product of around 8MHz, it does manage to maintain this comfortably up the 3MHz limit of my function generator. It is notable however that whereas the NE5534 setup clearly struggles at higher bandwidths, the DC imbalance between the inverting and non-inverting inputs is stable at 0mV. The DEXA however shows a high, frequency dependent drift in this imbalance, up to about 20mV or so. This again is bad news with respect to the TDA1541's 25mV compliance.
Differential Input Stage Matching And Temperature Drift
As would be suspected, further measurements show that the imbalance between the inverting and non-inverting inputs of the discrete DEXA is not only far greater than that of an IC opamp, but also very sensitive to temperature variations. In fact, although the DEXAs have a DC offset trimpot (tiny, and sensitive!), it is proving very hard to get the DC level of the DAC current output reliably close to zero Volt, especially when playing some music.
I know from experience that the CD80 internals run quite warm, and bringing up the internals to the operating temperature makes the DC offset on the DAC's current output drift by a good 15 to 20mV. Effectively this is a result of both the I/V feedback topology around the first DEXA with poorly matched differential inputs as well as the DC feedback servo doing its best killing the DC offset following the second DEXA. There are two DEXAs used more or less "in series" and unfortunately it appears the temperature sensitivities of both DEXAs in this particular configuration are additive rather than cancelling.
Combine that with the very sensitive trimpots (also a sign of future drift & stability problems), the frequency dependence of the imbalance, plus a 3mV or so ripple on the DAC's current output when playing a 20kHz 0dB signal level, and it soon becomes evident that keeping the DAC output reliably within 25mV compliance is going to be hard.
The DEXA Topology
These revelations have made me rather curious about what this “novel” DEXA circuitry actually looks like. Again, a schematic of the DEXA internals is not available, unlike for many other opamps (at least a search on the web for the circuit diagram was unsuccessful). It looks like we are going to have to do it the hard way!
The DEXA PCB is very small with SMD components. However, with a magnifier and a good light source it is possible to spot the traces on both sides of the PCB. Quite a few hours later I finally have a drawing of the PCB layout, and some time later in front of me lies a complete drawing of the DEXA circuit. This was not an easy exercise! Out of respect for NewClassD’s IP and as I don’t accidentally want to fall foul of any copyright legislation I will not reproduce the circuit here.
However, what I will reveal is that the design features a degenerated complimentary differential input stage employing four transistors. As this would ideally require the use of two matched pairs of transistors, this may help explain the relatively high imbalance and temperature sensitivity between the inverting and non-inverting inputs.
Further Considerations
The observations so far suggest that these DEXAs aren't quite the "drop-in" replacements they are claimed to be, at least not for a CD80. My guess is that this would be an issue with most TDA1541A based DACs, especially when employing a DC feedback servo. Furthermore, with the poor matching of the differential inputs of the I/V DEXA, the de-emphasis filter (part of the I/V stage) may also miss its target correction filter characteristic (I can't easily test that though).
Finally, the CD80's DC servo contains a couple of polarized electrolytic capacitors just outside the DAC output, feeding to ground via a resistor. As per the original design these are happy having virtually 0mV permanently biased either way when using the stock 5534s. However, with the DEXAs this could easily be up to 30mV reverse bias or so, depending on the operating temperature. In theory the caps should survive that, but I would nevertheless feel a little nervous having polarized electrolytic caps permanently reverse biased to that amount. Therefore, if the DEXAs are left in the I/V stage I would certainly replace these caps with bi-polar versions.
Rolling Only The Buffer Opamps
So, now what? Obviously one option is to keep the 5534 opamps in the first I/V stage, and use the DEXAs only in the second buffer opamp stage. Two DEXAs pulled and TI NE5534AP I/V opamps back in. O-scope tests show this configuration to be unconditionally stable as before, but the DAC outputs are now at a reassuring stable 0mV (well at least less than 1mV), even when playing a 0dB signal or any music. Heating the internals up makes no measurable difference whatsoever. Good.
Adjusting the trim pots on the remaining two DEXA buffer opamps to get the output of the preceding 5534 I/V opamps close to 0mV (feedback circuits are funny aren't they), and we are ready to go. This trim isn't really necessary as the 5534s are quite happy running with a DC offset of a few mV on their outputs, and the signal between the 5534s and DEXAs has no (practical) compliance limitations. It is just a bit of "tidying up" really.
Much Better
OK, this is better. Much better. The "coherent" sound, for lack of a better way of describing it, and PRAT are back. But (and this is even better news) the somewhat "grainy" treble of the stock CD80 has given way to an equally detailed, but more refined sound. The imaging is now also a tad better than with the unmodified CD80.
After a few days running-in time, things have only improved further. The sound is more transparent, has better imaging and more depth. The stock CD80 does have a subtly "warmer" sound though, but I would rather suggest the DEXA modified one is more neutral, especially when compared to the SA11S3, which it now comes a little closer to than does the stock CD80. The only "issue" is that it has made the Sennheiser HD800's even more unforgiving with respect to poor recordings. Good recordings however sound better than ever. As I said, the Marantz CD80 has not yet become a Marantz SA11S3, but it is getting closer now.
Rolling The I/V Opamps For BB OPA627
The next thing to try is to use a well-matched IC opamp alternative for the NE5534 in the I/V stage. Despite their relatively high cost, I have seen the Burr Brown OPA627 recommended time and time again. It is unity-gain stable, and the CD80 supply voltage is just about right, so we give it a go.
The OPA627BP opamps were selected and ordered from Farnell as they have slightly better characteristics than the OPA627AP, but are virtually the same price. Both of these get you to about half the price level of a discrete opamp, so they are still more than ten times the price of the humble NE5534. As these opamps have FET inputs they do require ESD handling precautions.
Swapping them in place of the NE5534 is otherwise straightforward. These opamps are significantly faster than the NE5534, so again operational stability was verified using a test disc and oscilloscope as before, which did not reveal any problems. The current output of the DAC was running at an expected negligible offset (<1mV). Incidentally, the 22pF compensation capacitors were left in as, just like the DEXA, the OPA627 does not have compensation pin 8 connected.
Much Better Still!
The OPA627 has a reputation for sounding a little laid back, if anything. I have to agree. In fact I would say the overall balance is now pretty close to the stock NE5534 / NE5534 combination. Typical somewhat laid back but nevertheless refined and engaging Marantz sound. So, was this then a very expensive way to get back to from where we started?
Fortunately, with this OPA627 / DEXA combination there are important aspects of the sound that have improved significantly. The CD80 now sounds really, really nice indeed. Whereas the stock CD80 at times can sound like it is struggling a little, this combination sounds effortless. A well balanced, neutral yet engaging sound with great PRAT and not a hint of listening fatigue. Interestingly, with this combination the Sennheiser HD800 is now again a little more forgiving of less-than-ideal recordings.
Although this is still not up to the SA11S3 level, it does manage to capture a good slice of the SA11S3 "magic" compared to the stock CD80. Whereas with the stock CD80 you always hear that you are listening to a very good CD player with a pleasing sound, both the modified CD80 and SA11S3 at times manage to make you feel the musicians are standing in front of you, or that you are actually in the church with the choir. Philips' 1989 recording of Jose Carreras in Ramirez' Misa Criolla is a perfect example. The modified CD80 manages to resolve higher frequencies more cleanly, and some of the harshness of the stock CD80 is gone. The acoustics of the recording environment are better defined, church acoustics in Ton Koopman’s Bach Organ works for example sound more life-like.
In fact it seems that particularly low-level sounds like the background sound and reverberations in a church or castle hall sound significantly more life-like, this being achieved without the need for emphasis on treble (the frequency response is still flat).
Furthermore, on the stock CD80 voices (particularly "s"-sounds) can sound a little "smeared" as far as positioning is concerned, whereas both the modified CD80 and SA11S3 are able to pinpoint them more stable and accurately.
The SA11S3 does give the impression of a somewhat wider soundstage, and generally still sounds a little more refined than the modified CD80. However, considering that the SA11S3 is still between five to ten times the cost of a used CD80 + cap & opamp modifications, the latter is a very attractive package if you are after a good example of the SAA7220P/B & TDA1541A sound. In that context, starting off with a factory configuration CD80, the £50 (~$80) capacitor rejuvenation is definitely worth it, as well as the £100 (~$160) opamp upgrade if you can afford it. And a Marantz CD94MkII could well be an even better basis, if you don't mind modding a legend...
Conclusions
When rejuvenating an older CD player (or other audio component), the priority should definitely be on replacing the dried-out old electrolytic capacitors. Where they are in the signal path (and that is not always clear-cut!) you can put some better quality ones in, or at least match what was in there already. When it comes to the decision on whether to replace the critical opamps with either IC or discrete ones, I have found that this choice may well depend on whether the circuit, and especially the DAC, can tolerate the poorer input stage matching of discrete opamps. Off course you can always combine IC and discrete ones, and the OPA627 & DEXA combination tried here works very well indeed for the CD80 in this case study. Whether the result justifies the financial outlay of these more expensive opamps very likely depends on the particular audio component. If the design is poor to start with it would likely be a waste of money, but in case of the Marantz CD80 it is definitely a worthwhile investment.
UPDATE 25-Feb-2014:
The other CD80 now has had the same opamp treatment, with the same improvements resulting.
The next step will probably be to replace the headphone amp in one of them (currently a pair of NE5532s). There is plenty of space and I have a spare pot & headphone jack salvaged from a broken CD60, which are identical to the ones used in the CD80. I will probably design & etch a couple of boards for just the amp section of something like a Lehmann linear, or improved version of my own DIY effort from 1992. I just need to check the CD80's power supply has enough spare capacity for the slightly elevated currents required.
