Trivista 21 is an upsampling DAC, whatever format you feed it, it internally upsamples to 96KHz or 192KHz depending on the position of the back-panel switch. Upsampling is performed by the CS8420 chip which is a 1990’s signal processor running a rudimentary algorithm in integer that is sure to introduce all kinds of errors.
My approach is to minimise the CS8420’s processing by setting the back panel switch to 96KHz and to feed the Trivista only 96KHz audio. By feeding 96KHz audio into the CS8420 chip and extracting 96KHz audio from it, the (nasty) effects of the chip is minimised. Roon’s DSP function performs sample rate conversion on the fly and also allows A/B comparison quickly and easily. Both 44KHz and 88KHz material sound much better when upsampled by Roon to 96KHz before being fed to the Trivista. The reason must be that Roon’s upsampling is better than the CS8420. This is hardly surprising: modern software with sophisticated algorithms running at 64 bit floating point precision on a powerful computer is going to be far superior to a 20 year old DSP. We are talking primarily 44KHz here: if the source material is already 96KHz then there should be no difference.
It is not possible to repeat the same trick at 192KHz because Trivista does not accept 192KHz audio. I expect it’s possible to bypass the CS8420 entirely by a hardware mod and if you have done it maybe you can share your findings. The 96KHz setting gives the solid, in-your-living-room kind of sound that I like. According to the data sheet, the DSD1792 DAC chip’s distortion at 96KHz is 0.0008% but at 192KHz it is 0.0015% so it is not clear to me whether the higher sampling rate will be better. I do have another 1792 based DAC that does give a better sound at 192KHz in my system.
If your source is Roon or other software that can perform on the fly sample rate conversion, go ahead and try it – I think you will like the result.
In this final part of the series I will talk about the (in)famous TriVista tube buffer and why you should bypass it.
What! I hear you say. Bypass the tube buffer? But the tubes are the reason for Trivista 21’s existence, right? I thought so too. Let me take you through the effort and expenditure I took to reach that conclusion. Let’s talk circuits. In the DAC section above we already discovered the output from the third op amp in the filter circuit is ready for output to the preamp. But instead of being made available at the back panel this signal goes through a unity gain buffer that contains the 5703 triodes. The 5703 triodes are not wired as traditional (and good sounding) cathode followers. Instead each channel has a pair of 5703 which form a long tailed pair in a discrete op amp configuration that also includes 7 bipolar transistors. Below are the improvements I made:
a) Matching the input valves: Performance of any long tailed pair hinges on perfect matching of the input devices so I purchased twenty JAN NOS 5703 and hand matched 2 pairs using an AVO valve tester. I did find the original valves were not matched.
b) Matching the NPN and PNP transistors in the second stage and output stage: originals were not matched at all.
Despite all this effort at a formal A/B test with and without the valve buffer all the Society members preferred the bypassed output with its much better transparency, communication of musical performance and far greater width and depth. I would like to try the tubes in a better circuit configuration in future and for now it was decided to permanently bypass the tube stage and disconnect the HT transformer to further reduce noise.
Figure 7: Trivista 21 (re-engineered)
After formal A/B comparison in a reference system we have determined that the re-engineered Trivista 21 is fully competitive with current DACs from Chord, Naim, Merging, Lumin. The secret is to correct the mistakes and use the correct (not just expensive) components. One last problem: now that the tube stage is disconnected do we still call it a Trivista?
Now we get to to the heart of the project – the DAC section.
MF made some mistakes in the analog power supply and the op amp filters. Below I will show you how to correct these mistakes and unlock tremendous improvements. Trivista uses a Burr Brown DSD1792A 24/192 DAC in PCM mode. This chip was state of the art in 2003 and boasts 127dB dynamic range and 0.0004% (-108dB) THD+N, which is comfortably beyond the performance required for CD audio. DAC chips are all quite cheap (tens of USD) on their own but they need good supporting circuits to make them sound good. TI published a reference circuit in the 1792 data sheet and I will refer to parts of it as we proceed.
+5V DAC chip analog power supply. DAC chips need separate clean +5V for the analog power pins. So why did MF supply the DAC chip’s analog power pins VCC2L and VCC2R from the dirty digital 5V power rail? After analysis confirmed by listening tests I concluded this was indeed a mistake probably made when they designed the PCB. There actually is a clean +5V rail supplied by a 7805 regulator with choke filtering that takes power from the already clean +15V analog rail but it is only used for the DAC chip’s VCC1. Get out your soldering iron and connect the DAC chip’s VCC2L and VCC2R pins to the clean analog +5V rail. You have just doubled your Trivista 21’s sound quality. For further improvement replace the 7805 regulator with something better. In this critical application the difference between a fairly good LM340A and a state of the art (in 2017) LT3045 is clearly audible in terms of transparency, resolution and spaciousness. If you want perfection you can use 3 LT3045 to power VCC1, VCC2L and VCC2R but you would need to make a piggy back board. Finish off with a sprinkling of OSCON on the DAC chip power rails. Don’t use OSCON for the internal bias decoupling caps because they have high leakage current. You have now doubled the sound quality again.
+15 and -15V op amp power supply. Each rail is provided by one 78/79 regulator taking power from the +-24 volt rails. Each regulator supplies both left and right channels. Putting in separate regulators for the left and right op amps would have cost USD5 more in components so this kind of cost cutting by MF is surprising. Unfortunately, there’s not much that can be done without seriously cutting up the PCB so I have left as is and merely upgraded the regulators to LM340. At least there is room to put in decent electrolytics.
Op amp integrator and filter. MF uses the reference circuit and I agree 100%. Some hobbyists claim they can come up with a better circuit, if they are so good they can go and design the DAC chip as well. However here is where MF made another mistake on the PCB (but not on the circuit). For some reason MF uses half of a 5532 dual op amp in all six positions instead of the recommended single 5534. That in itself is not a problem and maybe they had a lot of 5532 in stock. But they failed to disable the unused half of the 5532’s in the correct way. They wired both inverting and non-inverting inputs to ground, which ensures the output is driven at open loop gain and will peg itself at the + or the – power rail at random. Thus the non-working half of each 5532 compromises the working half. Correction requires track cutting to reconnect the inverting input to the output to regain stability. While you are working on the op amps why not use something better? The data sheet specifies the 5534 loved by big Japanese electronics companies who are the main customers. But the same circuit appears in OPA1611/12 data sheet so TI engineers know better. I am using LME49710 and LME49720 but if you use AD797 I would love to hear from you. While you are in there replace those cheap resistors and capacitors in the filters with your favorite precision types. I went with 0.1% Dale metal films with 5 and 15 ppm/C and hand matched 1% Phillips NOS polystyrene. There’s not much space to put in anything bigger/better.
Output buffer added. The output of the third op amp is ready to feed into your preamp. In the Trivista 21 this signal is connected to the input of the tube stage and is not available at the back panel. I added an LME49720 unity gain buffer stage so I could bring this clean signal to some extra phono sockets on the back panel to facilitate A/B comparison with/without the tube stage.
Now let’s turn our attention to the main PCB.
Figure 3: Main PCB (original)
Decoupling capacitors – stick with ceramics but use low-loss non-microphonic C0G types
If you are reading this you are no stranger to capacitors. I will leave you to decide whether you are going to increase capacitance or stick with original values. I would like to offer the following suggestions.
a. High frequency decoupling: There are about 50 low quality 0.1uF disc ceramic capacitors providing power supply decoupling onto the ground plane. Replace these with 0.1uF ceramic C0G. Remember this is a digital board with a 50MHz clock so you need caps that are good to at least 1GHz. Save your plastic films for analog stuff.
b. Low frequency decoupling: There are about 25 Jamicon brand electrolytics ranging from 10uF to 1000uF. I would go for 105C rating since you don’t want to do this again in the next 10 years. I stayed with Nichicon UKA because some 3 terminal regulators are said to be unstable with OSCON on the output. If you use OSCON, please let me know how it went. Some electrolytics provide critical voltage reference to the DAC chip and I will talk about these in the appropriate section.
1) Main power supply – quality choke filtered PSU, leave as is
Figure 2: PSU PCB (Original)
Power supply fanatics will be delighted to find large chokes on both the low voltage and high voltage analog power supplies. These chokes are bigger even than the power transformers. There are 2 main reservoir capacitors for each rail and they are wired in C-L-C arrangement with the chokes for maximum rejection of ripple, hum and other noise. The 24V rails will feed the 15V regulators for the op amps and the 90V rails will go through additional filtering before reaching the tube section. A 10V rail feeds the 5V regulator for the digital section and a 15V rail feeds the 12V regulator for the tube heaters; these rails do not have choke filters and probably do not need it. Best not to mess around with this superb design and just replace the old Jamicon electrolytics with new 105C rated components. I used Nichicon LGU at 50% bigger capacity and you should use your own favorite.
Trivista 21 circuit has several mistakes!!
I could not find any circuit diagrams on the internet, so I had to draw them myself by studying the circuit board. In the process I found several mistakes and shortcuts that absolutely prevent the Trivista reaching anywhere near its full potential. I was initially baffled and tried hard to convince myself that they were intentional. But after exhaustive engineering analysis backed by listening tests I had to conclude that they were indeed mistakes. I will highlight these in the appropriate section and show how they can be corrected.
Let’s get started with something all of us can see and that is the build quality. Trivista 21 build quality is excellent. Look at those beautiful double sided fibre glass PCBs with wide tracks and extensive ground planes. The wiring is terminated on the PCBs via proper termination pins and all soldered wiring is stress relieved with heat shrink. The last time I saw something this nice may have been during my days in military electronics.
What follows is quite a long article as I take you through each section and explain what I discovered.
Can the 2003 Trivista 21 be modernised to compete with today’s (2017) reference DACs? Enthusiasts have tried putting in expensive capacitors and op amps and some even offer their services for a fee. I applaud their efforts and generosity in sharing their work on the internet. But I wanted to go much further. When you peer inside the Trivista 21 you realise it is an extremely serious piece of equipment that demands an equally serious engineering effort to modernize it. This is not going to be a case of just swapping capacitors – this is going to be a total re-engineering. We will have to reverse-engineer the circuit, identify and correct all the weaknesses and deploy our components budget where it matters. With some work and USD 500 of components the Trivista 21 can emphatically withstand A/B blind testing against the very best DACs available today (2017) at any price. Provided you also bypass the tube section entirely!
To be continued.