I’ll admit that I get as much pleasure and satisfaction from designing and building audio gear as I do from listening to my system (well, almost anyway). It’s particularly rewarding when the sound quality of a new DIY component exceeds my expectations. I’ve had mixed results on that measure, but I am learning from my mistakes which I find worthwhile. I had been thinking about building another power amp as my next DIY project, but I’ve decided to wait to see how my 300B SET amps sound on the new speakers I am working on before I do that. I’m still waiting for the cabinets for the speakers, which I’m hoping will be shipped to me soon. A few months ago, I stumbled across a DIYAudio thread discussing a differential preamp design by Allen Wright. Allen passed away about ten years ago, but spent the last couple decades of his life perfecting his tube preamp design which was sold through his Vacuum State Audio business. The final incarnation was the RTP3D. Allen was fairly forthcoming regarding his design decisions and published the schematics and many of the design specifics of most of his products. The RTP3D received some very favorable reviews as a state-of-the-art preamp, but it’s not particularly well known in the US since Vacuum State is a small Swiss company and the preamp was quite expensive (~$25K). This thread will be a bit technical, but for those of you that are interested in learning more about the inner workings of a high-end tube preamp, and of course those DIYers that might be interested in building this beast of a preamp, I hope you will find it a worthwhile read. The RTP3 is a pretty complicated design with five dual-triodes for each channel’s line stage, and a sophisticated power supply. I am not trying to build an exact clone of the commercial product. For one thing, I am not going to implement the phono stage since my system is digital only. In other ways, my implementation will be a bit more sophisticated since I’ll be incorporating a microcontroller to control a color TFT display and provide remote control. The microcontroller will also handle power-up sequencing as well as controlling input selection, output muting, and a relay-switched attenuator. There are also components and transformers used in the commercial product that are no longer available which has necessitated some design changes. And some newer parts (such as the VCap CuTF caps and AN Silver Tantalum and Silver Niobium resistors) which I believe can further improve the sound quality. Allen originally built the RTP3 with all point-to-point wiring, but later switched to using PCBs. There are pros and cons to both approaches, but I’ve decided to build my preamp with PCBs. I am designing a total of eight custom PCBs, as well as using a few from other suppliers. The latter group includes the high-voltage shunt regulator boards used for the B+ regulation. This board was designed by DIYAudio contributor “Salas”. I am also using commercially available boards for the microcontroller and TFT display. The preamp will be built as a two-chassis implementation. The first chassis will contain the front-end power supplies, microcontroller with front panel controls and display. The second chassis will contain the analog circuitry and second stage power supplies. This project is going to take me a while given the complexity of the design. It will also be a fairly costly build, so this will let me spread the expenses out a bit.
RTP3 Line Stage One of the key reasons that tubes are popular among audiophiles is that they are inherently more linear than solid state gain devices. Triodes, in particular, allow voltage and current gain stages to be built with little or no negative feedback and still achieve reasonable results. Negative feedback reduces noise and distortion at the expense of shifting the distortion profile toward higher order harmonics and increasing distortion when the circuit starts to clip. Allen felt that the best results could be achieved with a zero feedback design, so he focused his efforts and using circuit topology to achieve the best performance without resorting to the “crutch” (Allen’s word) of negative feedback. The key elements he used to achieve this were triodes with good inherent linearity, a fully balanced, differential design, low-noise constant current sources, cascoding on the amplification devices, and a very robust power supply. I’ll go into more detail on these elements in a future post. For now, I’ll just give a basic overview of the amplification stage. The RTP3 has two stages - a differential amplifying stage (using two 6922 dual-triodes) which can be adjusted for between 10 and 30 db of gain (I’m going to set it up at the low end of this range), and a balanced cathode follower stage (using three 6922 dual triodes) which provides a robust drive capability and a relatively low 100 ohm output impedance. Both stages use cascoding, and the second stage uses an additional triode (per phase) for an active load. A negative 25V DC supply provides a negative bias for jfet current sources for each stage. And separate current regulators (five per channel) are used for the heater supplies of each tube. Even though it has zero negative feedback, Allen was able to achieve signal to noise of 122db with harmonic distortion (at 10V output) of 0.004% 2nd harmonic, 0.003 3rd, 0.000% 4th, and 0.001% 5th. Maximum output is 28V RMS. Those are very respectable specs even for a high-feedback solid state design, but very impressive for a zero feedback tube design.
That's a lot of 6922 tubes in the preamp. Do you already have a stash of good sounding NOS 6922 tubes?
The problem with the 6922 tube is that the good NOS versions are all gone. So we're stuck with the lesser NOS tubes or the new production. Even 10 years ago the good NOS 6922 tubes were around $200 each. Now they're gone. I have a couple of hybrid headphone amps that use the 6922 tube. I settled on the new production Genalex Gold Lion E88CC/6922 tubes for my amps. The Gold Lion tubes sounded better than the new production JJ tubes or the new production Electro Harmonix tubes. The Gold Lion and Electro Harmonix are made in Russia. The JJ are made in Slovakia. China isn't making a 6922 tube. And it doesn't look like Western Electric USA will be making a 6922 either. So Russia and Slovakia are our sources for new production 6922 tubes. And I'm not confident Russia will be a reliable source considering the invasion of Ukraine what they're doing right now. So I've been looking for alternative tubes that are compatible so I can keep my amps working and sounding good. I found that alternative with the GE JAN-5670W tube. It's not a direct replacement for the 6922. It requires an adapter to work in a 6922 tube socket. And it also has a 0.35 amp filament current draw compared to the 6922 which has a 0.3 amp filament current draw. If the circuit can handle the additional filament current then it can work in the amp. Another issue is that the 5670W can be more prone to noise due to more gain and filament noise. And surprise to me, the 5670W tube actually sounds better to me in my amps than the Gold Lion 6922. It costs less, is currently available, and sounds better. I like the 5670W tube so much that I bought enough of them to keep my amps going for more than a decade. iFi Audio was selling the 5670W as a 6922 replacement several years ago (they used the 5670W in their iFi Pro hybrid headphone amp). iFi Audio made a marketing paper promoting the 5670W as a 6922 replacement. Here's a pdf link to that marketing paper. iFi added a capacitor to the heater pin in their adapter to filter out some of the potential noise. If the RTP3 can handle the additional filament current and manage the additional potential noise then the 5670W tube could be a good option for that amp.
Yes, you are certainly right about the dearth of NOS 6922s. There are still some to be found, but pretty pricey. Thank you for the recommendation of the 5670W. I wasn't familiar with that tube. Mr. Wright recommended the Electro Harmonix tubes and these are what shipped in the commercial product, which is why I think I will start there. The commercial units received very high praise. I'm sure there is room for further improvement using nice NOS tubes, particularly in certain positions such as the differential LTP tubes, but I would like to hear how it sounds with the EH tubes first. I'm designing the heater supplies to be able to deliver at least 500mA to provide some headroom. The EH tubes can require as much as 335mA according to the data sheets. So I could certainly try the 5670W. This tube has a considerably lower transconductance than the 6922 so wouldn't be a good fit for the cascode positions, but might work well for the other tubes.
Analog Enclosure As mentioned in the first post, the preamp will be built in two enclosures. The inputs and outputs, attenuator, and line stage will be built in one enclosure, and the power supplies (first stage) and control circuitry (including front panel controls) will be built in a second chassis. This post will be about the first one. The two enclosures are fairly large, at approximately 480mm (19”) x 380mm (15”). This is about as big as I can fit on my rack. I haven’t figured out the exact height yet, but I’m shooting for around 4.5”. The analog circuitry will be built using two custom PCBs, duplicated for the two channels. One PCB will include the input and output connectors, input selection relays, output mute relay, stepped attenuator, and an opamp-based balanced to single ended converter to drive the single ended RCA output which I’ll use for my subwoofer. The second board contains the actual line stage circuitry, five current regulators for the tube heaters, a +/- 12V regulated supply for the opamps, and a -25v regulated supply for the current source bias. I’m planning to use VCap CuTF capacitors for the final decoupling caps and these will be mounted directly to the chassis under the input/output/attenuator board. These caps are daily large at 40mm diameter and 67mm length. In between the two channels are two Salas SSHV2 shunt regulators which provide the final B+ regulation for the two RTP3 line stages. The shunt MOSFETs will be heatsinked to a 10mm thick panel separating the regulators from the analog circuitry. I have preliminary designs done for all the custom PCBs except the Arduino Shield. But I still want to do some more review before I have the two described here fabricated. The image below shows how the boards will fit in the enclosure. The brown rectangles are the ends of the VCap CuTF caps used for the final decoupling (2 per channel).
My Conrad-Johnson Premier 17LS2 uses a quad of 6922 tubes. They are configured as one monolithic triode, so just one 4-triode composite gain stage per channel with a relatively low impedance output, no follower needed. I've used a variety of tubes in it, and the guy I bought it from included some nice ones too. I think right now it has some Amperex 6DJ8 Holland A-Frame types, which are pretty cheap in the wild. May not be suitable for a phono preamp 6922 substitute, but sound great here in a line amp.
That's an interesting approach. I've often seen two triodes paralleled to get more current capability, but I don't think I've ever seen four. Do you have to carefully match all the triodes to prevent current hogging? The RTP3 design will work ok with 6DJ8s since it doesn't really push the tubes very hard. I suspect they may not last as long though. I have an Amplitrex tube tester, so I may buy a bunch of used tubes off ebay or etsy to try to find some in good condition. It's typically possible to buy large lots of untested used tubes at bargain prices. This is what I did for the driver tubes in my 300B amps and I found about 25% of them to be pretty good.
Power Supply Enclosure The power supplies will be implemented completely separately for the two channels. Each channel will have seven different voltages, with first stage regulation in the PS chassis and second-stage regulation close to the analog hardware in the second chassis. The main B+ will use passive CLC filtering (followed by active shunt regulation in the second chassis), but all the other supplies will have two stages of active regulation. The B+ will use tube rectification followed by a CLC filter implemented using all film caps and a 17H Lundahl choke. The other supplies include three DC supplies for the tube filaments, one referenced to the B+ ground, and the other two raised to 125V and 215V. Separating these supplies allows the filaments to be at the optimal voltage for each tube in the line stage. The first stage regulators will provide about 10V DC. A second set of current regulators are implemented close to the tubes in the analog chassis. Each tube will have its own heater supply. Another two supplies are used to generate plus and minus 15V (for first stage regulation) which are used to provide power for a balanced to single-ended converter/line driver for my subwoofer output (my sub amps can only accept RCA cables). The final supply provides a -25V bias for the current sources used for the differential LTP (long-tail pair) in the first stage and the cathode followers in the second stage or the preamp line stage. I am having Toroidy produce two custom toroidal transformers - one of each for each channel. One will provide the AC voltage for the B+ supply, along with the rectifier filament supply, and for the current source bias. The second one will provide three separate center-tapped secondaries for the three heater circuits, and the two secondaries for the +/- op amp supply. Three separate circuits are needed for the heaters because the heater supplies need to be within a certain range of the tube cathode voltage. So one is referenced to ground, but the other two are elevated to higher voltages (~125V and ~230V). I’ve designed three custom PCBs for the power supply chassis. The square shaped board in the middle rear is a power controller which provides a soft start for each of the two transformers for each channel. This allows the tube heaters to be turned on separately so the tubes can be heated up before the B+ is applied. This board also provides a 12V trigger output. There are a total of eight (4 per channel) identical DC regulator PCBs for the three heater circuits and the op-amp positive supply, and four (2 per channel) identical negative DC regulator PCBs for the CCS bias supply and the op-amp negative supply. Since the op-amp current requirements are quite low, smaller heat sinks can be used for the regulators, so the positive and negative boards can be stacked. The final PCB is the B+ supply. This uses a 6CA4 rectifier tube along with two UF4007 rectifiers to provide a full bridge rectifier. This feeds a CLC filter using ClarityCap TC4 film caps and Lundahl chokes. This board also generates the bias voltages for the elevated heater supplies. The front part of this enclosure will house the Arduino control logic. This will be very similar to the controls and display I used for my previous Aikido preamp project. But this time I am doing a custom PCB for all the wiring so that I can easily use pluggable connections. A ribbon cable will connect to the power controller board and then to a rear mounted D-shell connector. A standard VGA cable will be used to provide the control connection between the two enclosures. I’ll use AC-DC power modules mounted to the rear of the front panel to provide the 5V and 12V power required by the controller logic. These will be powered up whenever the rear panel power switch is on so that the remote control will allow the preamp (and amps using the trigger) to be turned on and off.
I don't think the matching is that critical, though I do generally try to match channels by swapping the tubes around since it's a no feedback design. The 17LS2 is based on the original ART design introduced in 1996, which used five 6922 dual-triode tubes per channel, so a composite of 10 triodes, and was a dual mono package in separate left and right channel boxes. Amazing preamp, but very expensive. The Premier 16LS was the simplified single-box 6-tube version in 1998, while the later Premier 17LS was a single box 4-tube version in 2000. After that, they moved to their current GAT topology, which just uses one dual-triode 6922 per channel and a MOSFET buffer. Premier 17LS monolithic gain stage ...
I'm a bit of a fan of the humble EH6922, at least compared to the other new production 6DJ8 tubes. If you find they do a decent job and don't feel like going crazy with the usual NOS suspects, then you'd do worse than buying up a batch of Soviet era 6N23Ps on eBay. They can be purchased in the tens, fifties, or hundreds sometimes. Although the 6N23P is bested in many areas by other NOS tubes, I personally haven't found one that is as much of a satisfying all-rounder. The Electro Harmonix is, I believe, actually a modern 6N23P and they both have a similar sound, but the Soviet tube just does it "better". They both idle at a higher current than other 6DJ8/6922 tubes do when cathode biased. As for @Ham Sandwich's suggestion of the 5670, I agree it's worth looking into. I used the Soviet version, the 6N3P, with an adapter in my ANK DAC and preferred it to the 6N23P in all respects. The top end especially was more natural and open sounding. Apparently the JAN 5670 sounds even better. Funny... ...I thought your other projects were ambitious... edit: added in the important "don't"
Just placed an order for 50 of these tubes...pretty cheap. This will keep me busy for a few hours with my tube tester.
I wish you the best of luck on this one. A very ambitious project, but soooo much potential. In the little bit of reading so far on his stuff. I really like his concepts.
Thanks. I'm pretty excited about this project. There is a thread on What's Best Forum Vacuum State RTP3D preamp including a discussion of a recent custom build someone is doing, where I got some ideas of other tubes to try. Specifically, the NOS Mullard/Amperex 7308 which is a low-noise variant of the 6922 and seems to be available from a few vendors, and the PCC88 which is a 7V version of the 6922. My design uses current-regulated independent supplies for each tube and is adjustable from 250mA to 350mA and can supply voltage up to about 8.5V, so can easily accommodate these tubes. I got some test PCBs today for the positive and negative DC supplies, the power controller, and the high-voltage supply. These are cheap PCBs that I made to make sure the circuit works ok, and to check clearances, hole sizes, etc. Once I confirm that these are working, I'll order higher quality boards (heavier copper, thicker FR4, ENIG plating) that I'll use for the final build. I've decided to do my own PCB layout for the Salas shunt regulator so that I can add a large film cap (the same ClarityCap TC4 that I am using in the power supply chassis) on the input of the regulator. This will also allow me to move the current-limiting series resistor to the side of the board so that it can be heat-sinked to the chassis panel.
B+ Shunt Regulation The main high voltage supply in a tube amplifier is referred to as the B+ supply. Early tube amplifiers were powered by batteries, so B+ refers to the positive battery voltage, and even though most tube amplifiers haven’t used battery power for well more than half a century, this term has stuck. Allen Wright believed very strongly in the influence of the power supply on sound quality (as do I). He felt that the quality of the power supply (for all aspects of the preamp) was at least as important as the amplifier stages themselves. And if you look at the top preamps on the market (whether tube or SS), you’ll see that a great deal of attention is paid to this part of the design. Alan did a lot of experimentation with power supply concepts and found that shunt regulation fed with a current source delivered the best sound quality. He developed his own shunt regulator design which he called his Super Regulator. There are two basic approaches to linear voltage regulation - the series regulator and the shunt regulator. By far the most common regulator is a series regulator. They are simpler to implement than a shunt regulator and much more efficient. They work by using a series device (bipolar transistor, mosfet or tube) to “throttle” the amount of current that is passed to the load. A shunt regulator, on the other hand, assumes that more current than required is available, and shunts the unwanted current through a device to ground. I’ll use a more concrete example to explain how these work. Let’s assume we have a water bucket with a hole in it. The amount of water that flows through the hole represents the current requirements of the amplifier, and will vary based on the signal and load requirements. The goal of the regulator is to keep the water level in the bucket constant. A series regulator works by controlling the flow of water into the bucket to keep the level constant. So it will never deliver more water than is necessary to maintain the level. A shunt regulator is like having another spigot on the bucket. It works by assuming that more water than is needed is flowing into the bucket, and the shunt regulator drains the excess. So clearly the shunt regulator is more wasteful. We are pouring more water (current) into the bucket than we need, and throwing the excess away (turning it into heat in the case of our electronic shunt regulator). So why would we ever want to use a shunt regulator when it is so inefficient? If the preamp was driving a purely resistive load, there wouldn’t be much benefit, but in a real preamp, we are driving cables and amplifiers and internal circuits that do not represent a purely resistive load. The load stores and releases energy. When energy is released by the load, it must be absorbed by the preamp. This can momentarily cause the preamp’s current requirements to become much less or even negative. This can play havoc with series regulators because they are designed to operate with at least a minimal positive current flow. A shunt regulator, on the other hand, can just increase the amount of current that is dumped to ground (burning it off as heat in the regulator). The shunt regulator can therefore maintain a constant voltage under real-world conditions more effectively than a series regulator.
Channel Separation To achieve the best imaging and soundstage, channel separation is paramount. Any crosstalk between the two channels will compromise these effects. You probably have noticed that the top preamps are often built with two, three, or even four separate chassis to minimize interaction between the power supply and the analog circuitry, and between the two channels. To achieve maximum separation, the power supplies, and even the grounding, has to be completely independent between the two channels. In my own experience, a two chassis implementation - with the transformers, rectifiers, and first stage filtering in one chassis and the analog circuitry with final regulation/filtering in a second chassis - is a good practical approach to delivering optimal performance. This assumes that the power supplies and analog circuitry are completely separated for each channel, including the ground connections. In this preamp build, I will do just that. Completely separate power supplies and grounds between the two channels. Although I will not resort to using separate power cords for the two channels. That seems like a step too far (at least for a preamp).
Thanks for the excellent overview of regulators, especially the easy explanation of the shunt kind. I've never really found the electricity to water analogies to be very helpful but in this case it was simple to follow. It looks like the complexity of shunt regulation designs can be quite variable? I haven't investigated the Salas in depth but it looks like there's quite a bit going on with multiple transistors, which is an interesting counterpoint to the shunt in the top end ANK kits, it just being a large power resistor. I have to admit, while reading the post that followed the above, I was quietly thinking to myself... I hope he uses two power cords
B+ Supply (part 2) For my preamp build, I am building a separate power supply for each channel. This includes supplies for tube filaments, op-amp supply (for the subwoofer outputs), and current source bias, in addition to the B+ supply. Using completely separate supplies allows separate grounds for each channel, so there is no electrical connection between the two channels other than the AC mains connection. The B+ supply for each channel has the following subsystems: Mains RFI filtering. Soft-start. Transformer. Snubber circuitry Hybrid rectification CLCRC filter Current limiter Shunt regulator I’ve implemented three separate custom PCBs for B+ circuit. The first is a power controller. This performs four functions. The first is to provide a RFI filter on the mains signal. This is done with X and Y rated capacitors that shunt high frequency noise before it can cause problems in the audio circuitry. The second function is to provide a separate turn-on for the heater circuits and the main power. This allows the tube filaments to be energized before the main B+ is turned on. Both the tube filaments and main B+ power have a soft-start circuit. This places several NTC thermistors in series with the AC mains connection for a couple seconds until the transformer circuit is energized, at which point the thermistors will be bypassed. This significantly reduces the turn-on surge current and reduces the stress on the transformers, rectifiers and filter capacitors on initial power-up. This functionality is implemented using relays that are controlled by the Arduino processor (that also controls input selection, attenuation, etc.). I’m using two separate custom Toroidy transformers per channel - one for the heater and op-amp circuits, and one for the B+ and CCS bias. This allows the two circuits to be turned on separately. The B+ current requirements for each channel are roughly 40mA (not including the extra current that will be wasted by the shunt regulator). I’ve spec’ed the transformers at 200mA (AC), which provides a reasonable margin to prevent the transformers from being stressed. These transformers (the ones for the B+ supply) will also provide the rectifier heater current as well as the negative bias supply for the current sources. I’ve chosen to use Toroidy transformers because I’ve been happy with the previous transformers I’ve purchased from them and because they are well shielded both electronically and mechanically. Toroidal transformers generate less EMI so other circuits are less likely to pick up any spurious noise.
B+ Supply (Part 3) The transformer outputs feed the next custom PCB which has the snubber circuit, rectifier, and the first part of the filtering (CLC). A transformer feeding a rectifier will have a fair amount of overshoot (and ringing) because the transformer is an inductor. This overshoot causes excess noise which propagates into the rest of the circuitry. It’s possible to design a tuned RC filter on the transformer output to minimize this overshoot. To achieve the best results, this circuit has to be tuned for the specific transformer. Fortunately, a DIYAudio contributor (Mark Johnson) has created a simple circuit to determine the optimal snubber circuit values for a specific transformer to minimize the overshoot and ringing. This circuitry is implemented on the high voltage power supply board shown below. The PCB includes the rectifiers and the first two capacitors in the CLCRC filter. I’ve decided to use a hybrid rectifier using a tube dual rectifier for the positive current and soft-recover SS diodes for the negative. The main advantage of a tube rectifier is a slow power-up and softer turn-on/off. The hybrid circuit provides most of the benefit of a full bridge tube rectifier while using only a single tube. There is no substitute to using a large inductor to filter ripple. The amount of ripple reduction from a choke followed by a cap is pretty astonishing, particularly from those of us that are used to dealing with SS circuits and relying on capacitors alone for filtering. I’ve decided to implement the B+ supply without any electrolytic caps. Electrolytic caps provide a lot of capacitance for the size and cost, but also have much more limited life and much higher dielectric absorption, and are generally thought to reduce the sound quality. With a large choke, it’s not necessary to have huge capacitors to eliminate ripple. The CLC circuit I’ve implemented cuts ripple down to a few millivolts before feeding the shunt regulator. The first capacitor after the rectifier is fairly small, but can be adjusted to provide the optimal voltage to the shunt regulator. Small changes in the first cap can be used to adjust the output voltage, but using too large a capacitor will increase the surge current through the rectifier which could put too much stress on the rectifier tube. So care must be taken in choosing this cap value. I’m using a Lundahl 17H choke followed by a ClarityCap TC4 50uF film cap. These are both high quality parts and should provide very hiqh quality power to the next stage in the power supply. This board also includes voltage dividers to generate the voltages to bias the tube heater supplies so that they don't exceed the specs for the tubes. The heater supplies must not be at a voltage that is too positive or negative relative to the tube cathode or this could cause the charge to flow to/from the filaments when it's not supposed to, and ultimately could cause damage to the tube.
Sounds like a really well thought-out project! You're working at a pretty high design and component level, I'm interested to see how it all turns out. Bet it does sound good I'm a big fan of shunt regulators and all-film capacitor designs too. When I designed and built my DAC in the late 90s, I developed a simple shunt regulator that didn't require any added series element, which is usually a resistor or current source, instead I regulated across the choke impedance. The regulators also used an isolated lithium battery reference, so no feedback from the input or output voltage to the voltage reference as in normal designs. I used FET input opamps that were selected just for that purpose, so they didn't have any input leakage from the battery, and the lithium cell lasts about as long as the shelf life specification. The shunt element is a power MOSFET, and the configuration I chose is scalable to higher voltages without any significant changes (I used the design in a power amp too). In the schematic below, the circuit on the left is the +/- 18V regulators for the DAC analog section, while the right circuit after the 300 ohm series resistors provides +/- 5V to the D/A convertor IC, and also to the regulator opamps...
@Davey - Looks like a clever approach. I bet it works well. What kind of DAC are you using with this?
