An Interview With Experienced Digital and Analog Designer Richard Dudley

1) Can you give us an overview of your background, how it relates to audio reproduction, and where your greatest interest currently lies?

I got interested in electronics as a child (I think around 10 years old) when I had a class teacher who had an electronics set in the corner of the room. There was one time a week when we could all choose which of the class ‘toys’ to play with, the electronics set was the one which grabbed me, not sure why. I can remember trying to work out, with a transistor, how the current could ‘decide’ which way to go between the two other terminals (it comes in on one of the three legs). I never figured that out! A best friend at school a year or so later had a ‘Denshi block’ electronics set – where all the components are encapsulated in plastic boxes with metal terminal lugs on the outside, these blocks pressed into a grid. I investigated and found I probably couldn’t afford so luxurious a kit so I settled on a cheaper ‘Philips’ set which I calculated I could buy if I sold my collection of Lego. Lego had been my primary toy for as long as I could remember so it was a real watershed for me to sell that off and re-invest in electronics. I’ve not looked back since!

As a secondary school boy a few of my colleagues were also interested in electronics so we egged each other on in the projects we’d undertake although none of mine had much connection to audio. I had a couple of articles published in magazines and earned a little pocket money through that. In the sixth form I got a job at a local electronics shop on Saturdays (later it grew into a much bigger enterprise Watford Electronics )

The job helped with my electronics knowledge and also provided parts at a discount. Seeing as by this stage electronics was my life-blood it made sense to me to apply for an EE course at university where I picked up some theory to back up my years of practice and experience in the hobby field. It was at uni that I first got bitten by the audio bug – I had a ‘music centre’ which accompanied me to and from my lodgings each term of my first year but my heart was really set on acquiring a separates system, fueled by poring over the magazines ‘What HiFi?’ and ‘HiFi News’. Using the earnings from my summer job I was able to fulfill my aspirations – one component being the now legendary NAD3020 amp.

For several years after I graduated my interest (and work) in electronics was quite separate from my pursuit of audio. This ‘Chinese wall’ started to dissolve when I undertook to re-design the PA system for the church I attended. It was such an interesting challenge that after I’d completed it I got into building myself a poweramp and modifying my CD player. This was around the time that Ben Duncan published his series of DIY pre-amp articles in HFN+RR so I owe much of my inspiration to Ben. I was able to return some of the favour in proof-reading some of his book on power amplifiers years later. I eventually left my job in industrial vibration monitoring and went to work as digital designer for a pro-audio company. Audio really had taken over my whole life – the year was 1989.

An alpha prototype of the DAC filter described in my Hackaday project’

My greatest interest nowadays is in asking and answering the question ‘How low can I go?’ – by which I mean how cheap can I make something while it still sounds good? I have long been of the view that audio kit is overpriced – but this is just an expression of the problem that all electronics made in Western countries is overpriced. (Jason @ Schiit recently found out he was being royally stitched up on the price of his AD DAC chips – this is a rather different, but nevertheless very interesting topic). Since I’ve been living in China I’ve had my eyes opened to just how cheap audio gear can be, from many hours shopping on Taobao. I’ve posted up some of the extreme bargains I’ve found to my blog –  diyAudio – abraxalito

I started out my current line of research by asking the question….”how much of my present level of audio reproduction is due to having sub-optimal speakers and how much is due to the electronics?’  As I’m not a speaker designer I decided to begin an answer to that question by tweaking some ready made audio components, which of course I bought on Taobao. The first one was a DAC based on the AD1955 – which I went crazy on modding. I was feeding it to cheap active speakers at the time for the rest of the system and fed from a QA550 SD card player. I found the precise topology of wiring layout in the DAC and speakers made a difference to the sound – star grounding sounding by far the best. But most DACs use groundfills because that’s what the manufacturers tell designers to do in order to achieve lowest power supply inductance. So I started to doubt whether manufacturers really had listeners in mind, they are really focused on getting the best numbers….

2) Can you tell us about how you view measurements in the context of audiophile grade audio components? Also how they tie into the marketing of these products?

How I view measurements to a large degree depends on which hat I’m wearing – design engineer or consumer. Wearing my EE hat measurements are an invaluable way for me to tell if my design in theory has worked out in practice. They’re a way to tie models to reality. Just as one example – on my present DAC design I have a filter which if not given the correct load impedance doesn’t give a flat frequency response (FR). The impedance, coming as it does from an active circuit is hard to measure directly but the frequency response is quite quick to check, so the flatness is a way of seeing if my theory for the input impedance of my active circuit is borne out in practice. At first it wasn’t, and by quite a large margin so the FR measurement was crucial feedback in the design process. I had to adjust my mental model for how the circuit really worked.

As potential buyer of an amp or DAC, measurements as they’re presently constituted are far less valuable. The manufacturers of course are likely to use them in the same way I do, to check there are no mistakes made in the manufacturing of the product, that the amp sold is indeed the design the engineers consciously intended it to be. As guarantors of audio quality though, the current crop of measurements is severely lacking.

Take for a start THD measurements. Nowadays even ‘objectivists’ will say they’re not particularly useful without knowing which harmonics are contributing to the number. This is a good start (Earl Geddes wrote a paper with a new metric proposed) but doesn’t go nearly far enough, mainly because the stimulus (a single sinewave) doesn’t look in any shape or form like music. Music’s ‘crest factor’ – a measure of the ratio of peak to average power will usually exceed 10dB and on classical music (the form I spend most time with) will be 20dB or so. For a sinewave this number is 3dB. How this makes a difference is that its the low signal-level performance of an amp (the now legendary ‘first watt’) that is most crucial – a single sinewave spends such a small fraction of its time exercising that range of output levels that its close to useless at characterizing SQ (sound quality). In the early days of audio, sinewave test tones were the only ones practical to produce, in the digital era though there is really zero excuse not to test amps with a high crest factor signal. Indeed Audio Precision has had this facility for a couple of decades but it has still to catch on. There’s at least one exception though, that being ‘Neurochrome’ who are showing multitone test results for their ‘Modulus’ range of amps. In part this is down to my low-key educational program on DIYaudio – the founder of this numbers-driven audio business hangs out there.

The Domino DAC – fully balanced with transformer output

THD measurements on DACs have turned into something of an ‘arms race’ amongst chip manufacturers for who can deliver the lowest possible numbers. I suspect that THD numbers have been part of the reason why sigma-delta DACs have upstaged the older multibit chips, as their vital stats look so much better, with immaculate-looking FFTs down to the lowest levels. Speaking of FFTs I won’t let them pass without a comment – there are artifacts which are easy to see on FFT plots so those are the things designers target for elimination. There are others (noise modulation being perhaps the most important one) which are much harder to spot due to the way the FFT operates. The monoculture of measurement tools (THD, FFT) hasn’t been a good thing in my view for subjective sound quality – we need only look to economics to begin to see why. In that field there’s a saying, Goodhart’s Law which goes ‘When a measure becomes a target, it ceases to be a good measure.’ THD and FFT are in my mind akin to the measurement tool which contributed to the last financial crisis, ‘VaR’ (value at risk).

3) I understand you have been working on a unique DAC design for a few years now. Can you explain your concepts in as much detail as you can without compromising any proprietary data?

I’ve been working to understand what the main contributions to how a DAC sounds over a number of years yes. The first DAC design I did was while I was working at SSL over 25 years ago – this was using the new ‘Bitstream’ Philips chips (SAA7321) – at that stage about the only slightly different thing I did was drive two chips in anti-phase and re-ordered their data so both phases of ‘left’ came out of one chip and two ‘right’ phases came out of the other. I didn’t evaluate the SQ particularly critically in those days as I hadn’t enough listening experience to build up a mental picture for how a good DAC should sound. What I did notice in these first generation S-D chips though was low-level ‘birdies’ during quiet passages. My next design was using the SAA7350 and in that device the low-level idle-tone performance had been improved. I took the view at that time that it was really low-level linearity that mattered in a DAC which was why I was fairly well convinced that S-D DACs were the future.

Fast forward to when I started my ‘hobby’ project, I kicked off with a latest generation S-D DAC chip in a relatively poor implementation from Taobao. I figured if it didn’t sound good it would give me a chance to explore how cleaning up the layout (as the manufacturer’s recommendations had largely been ignored) changed the sound. What surprised me was how much better I could make this sound with relatively simple tweaks. The really low hanging fruit for DACs I discovered was getting a good layout, which of course contributes nothing to the BOM cost whatsoever. I played around quite a lot with the output op-amps, at that stage I had no inkling of how sensitive op-amps were to their power supplies. I discovered that to get the best sound out of an op-amp in that position the feedback capacitance should be as low as humanly possible. Despite many hours spent hunched over my computer simulation I was never able to find a connection between the value of that capacitor and any changes in measurement (meaning THD).

A classA transformer output headphone amp

Having got a very satisfying improvement in the sound from that S-D DAC I was curious to learn what the attraction was with NOS DACs so I bought a very cheap toy TDA1543 board. I found it a bit dull, lacking ‘detail’ at first but its qualities gradually grew on me. I tweaked up the famous ‘NOS droop’ to get a flatter frequency response which helped but I also threw out the output RC filter. I then wondered how a better measuring multibit DAC might sound so I investigated the CMOS cousin of this chip, the TDA1545. I had been given samples of this back when it was first introduced but never put them in a circuit. Being super-lazy about building from scratch, the vehicle I chose for them was the Lite DAC-AH which comes kitted out with eight TDA1543s (they have the same footprint) and an op-amp for the output stage. What was of interest with this project was the impact of low-pass filtering between DACs and output stage. I found I was able to control the amount of ‘detail’ in the sound with a series of ferrite beads – the more beads I added, the less detail I heard. But as this ‘detail’ got quieter and quieter I found I was rewarded with improved tonal colours and more depth to the soundstage. It then dawned on me that the ‘detail’ I had missed on first switching over to NOS from my AD1955 was an artifact, not a feature.

The absence of ‘detail’ with the TDA1543 and its presence with the unfiltered TDA1545 made me wonder whether the reason the latter chip hadn’t gained such traction with audiophile DIYers was they hadn’t appreciated the ‘detail’ that arrives when the 1545 is directly connected to an op-amp. I reasoned that ‘detail’ was caused by RF upsetting the input stage of the opamp, the RF originating at the power supply of the DAC. The TDA1543 being a bipolar (referring to the types of transistors used internally) design doesn’t pollute its power supply to the degree of any CMOS DAC but it draws at least 10X the current because its ‘on all the time’. CMOS chips only draw power at clock edges but this characteristic modulates their supply to the detriment of the sound. Seeing as I wanted my design to use as little power as possible, I decided to go forward with the CMOS DAC together with some kind of low-pass filter to remove the ‘detail’.

The kind of filter I have used in my designs has changed a lot as I’ve experimented with multiple options. Because inductors in passive filters tend to be bulky things, as well as having relatively poor tolerances, for a while I was using an active filter. But not one based on op-amps, rather discrete transistors. Op-amps generally aren’t great for sound quality but I’ve not anywhere read about this weakness being tied to any objective characteristics. The key to why they’re bad seems to be they run in classAB so contribute noise to their supplies when driving any kind of load. To get the best sound out of any op-amp, don’t make it work at all hard. JFET input stage types are best because they’ll work with very high resistor values without becoming too noisy. If you use it with no smaller than 200k resistors even the humble TL084 can put on a good show. Nowadays I avoid them and use discretes in classA to maintain hygiene on my power supplies.

A very important aspect of a DAC is its I/V stage – the older generation of S-D chips (prior to ESS) like AD1955 explicitly tell designers not to use passive I/V – op-amps are de rigeur in manufacturers’ recommended schematics. DIYers though tend to like passive I/V in NOS DACs which is where I started out when I began playing with TDA1543. The traditional EE view is that passive I/V sucks because DACs have signal-dependent output resistance, so having a relatively high impedance load causes distortion. Whilst this is true without a doubt, I have doubts that its the reason for differences in sound between active (non-op-amp) and passive I/V. That’s because when using passive I/V I found differences in the power supply impedance made a difference to the sound, most markedly in the bass. I experimented with huge amounts of capacitance (I gave up around 1F, this not being made from a supercap which has too high internal impedance for the job) and always found more = better bass.

A completely off-the-wall power decoupling design for a classD amp IC’

I built some prototype DACs as ‘towers’ where the tower was built out of a hexagonal cross-section of capacitors, stacked vertically. But the time consuming nature of creating such dissuaded me from turning these DACs into published designs. Its just too impractical to parallel up hundreds of caps, not to mention the sheer bulk! I was pondering a solution for what seemed a very long time – until I happened to notice that the bass of a commercial DAC using opamp I/V wasn’t as much affected by adding capacitance. Op-amps present a very low impedance to a DAC when used in ‘virtual earth’ mode. Of course op-amps have other drawbacks – primarily in the higher frequencies due to supply noise I mentioned earlier – but I got to thinking that perhaps lower impedance to the DAC would be beneficial. A passive I/V resistor of lower value doesn’t help as lower values need higher gain, what’s needed is a true current-to-voltage converter. So I built one from a common-base transistor – this got incorporated in a modification to the commercial DAC I mentioned earlier – if your interest is piqued the discussion about this DAC is here:

Sure enough the common-base transistor did the trick of giving better defined bass without needing a hundred or more caps. I was another step along the path of reaching my goal of a pocket-sized DAC with audiophile sound.

The common-base transistor has the slight drawback that it needs a fair bias current in order to present a low impedance to the DAC, and its impedance changes slightly with signal current. I figured I could improve on it by using negative feedback around a common-base input transistor. This gave a lower powered solution with more stable input impedance which I first incorporated into a prototype I called the ‘Domino DAC’. It acquired this moniker because it has two sets of six LEDs in a pattern which look like a ‘double six’ on a domino. The LEDs were used as references for the active output filter. The design has no inductors so is nice and compact – being balanced though it needs output transformers which do tend to be rather bulky!

The final innovation – I call it that because I’ve not seen it applied in any commercial or DIY DAC to date – is to move the anti-imaging filter to a position upstream of the I/V stage. The advantage here is that no active circuitry is exposed to the ‘raw’ wideband output from a DAC which can easily reach into the 10’s of MHz. Lynn Olson verified the wideband nature of his PCM63 DAC’s output with an RF spectrum analyser so became convinced of the need for some filtering. My filters take his suggestion for rather gentle low-pass filtering much further and are still a work in progress – to catch up on the latest developments, pop over to my Hackaday project:

Audiophile-sounding DAC for almost no money