Dear Rob
What is a OP stage? I understand discrete stage is better than op-amp, could you explain why? As I understand the Hugo has no analog volume control, so the output from the DAC doesn't go through a preamp (like one of the competing products from Salisbury)
Also what is a pulse array dac? is it similar to Delta Sigma or the resistor ladder Dac? Is the sound of the hugo due to the filter or due to filter/dac combination? Also if you were to use this filter with a conventional resistor ladder DAC would it work?
Welcome to Head-Fi analogmusic, and I am pleased you are enjoying more musicality from your music with Hugo - which is what this is all about!
What is an OP stage?
OP is output, and it replaces rather poor OP stages within op-amps. When faced with designing the electronics of Hugo, I had no experience of designing headphone amps - looking into devices that supplied headphones, they were very poor. So I designed it as if it was a power amp (I've designed lots of those) and gave Hugo the ability to drive 8 ohm loudspeakers directly - which means lots of current - in Hugo's case I set it too 0.5A RMS. You will not get this current from op-amps or headphone drive chips, so I had to design a discrete amp. Now to get the best transparency there needs to be a single feedback path, so the discrete OP stage needs to be within the op-amp's global feedback path. Since the op-amps are very high gain bandwidth product devices (high speed), that meant designing a Class A OP stage with very low propagation delay, so that the circuit would remain stable. Now the op-stages in op-amps are pretty poor to awful, so when I got the first prototype I was very pleased at how good the OP stage sounded, and how much lower distortion was (particularly high order harmonics) - even when using the op-stage in DAC mode with easy loads. Indeed, I now use this arrangement all the time now, as it really improves the performance of the op-amp - that's why 2 Qute has it too. The OP stage is by far the weakest part of all op-amps and this is simply because one can use a decent Class A bias current, and very substantial OP transistors, so thermal stability is ensured. And yes, Hugo does not have an analogue volume control, so this means the analogue section is very simple (just 2 resistors and capacitors in the direct signal path). Simple analogue gives much more transparency.
What is a Pulse Array DAC?
This is not an easy answer, as its complex and of course proprietary. But firstly the history. I first started designing DAC's in 1989, when the first delta-sigma bitstream devices from Phillips came out - these were DSD 256 DAC's (or PDM dac's). Now they were quite musical, but had technical and SQ problems - but they had very good low signal performance, and analogue distortion characteristic (small distortion for small signals unlike R2R DAC's which have more distortion for small signals due to glitch energy and resistor matching problems - issues that are impossible to solve). The biggest problem was limiting of resolution - unlike PCM, where ultra small signals are buried in the dither and so perfectly preserved, with delta-sigma the noise floor is a cliff edge for low level signals - any small signal below the resolving power of the noise shaper is lost forever. To overcome this, I used 8 PDM noise shapers with different dither, and summed the output in the analogue current to voltage converter (I to V). This gave much better performance, but I knew that much more was possible. So I started creating my own noise shapers and DAC technology using FPGA that were just becoming available (1994 now). What I needed was much higher resolution so the noise shaper OP is 5 bits not 1 bit, and I ran the noise shapers at a much higher rate - 2048 times not 256 times. Running at a faster rate means that you have more permutations of OP, which translates to much better performance. Run a 5th order noise shaper at ten times the speed, you can get in the digital domain, up to 100 dB lower distortion and noise - that's a 100 dB improvement in small signal resolution, so running at much higher rates gives massive improvements in SQ and measurements. Twenty years on, and I am still the only silicon/FPGA DAC designer running as high as this rate - delta-sigma DAC's are still stuck at 256 times or below.
But changing from single bit to multi-bit noise shaping may throw the baby out with the bathwater. The primary benefit of single bit is that it
can (if you are very very careful) have zero small signal distortion, as there are no resistors to balance, as there is only one. With 16e Pulse Array, there are 16 PWM elements, and each element has on the long term exactly the same data, but instantaneously slightly different data. The benefit of the Pulse Array scheme is that when the elements are slightly different in value, it creates a fixed signal independent noise, and absolutely no distortion, but has innately higher resolution of 5 bits. That's why Hugo has (uniquely compared to other non Pulse Array DAC's) no measurable distortion, or any other artifact, for signals below -30 dBFS (see plots in previous posts). Additionally, because of the way the array is composed, master clock jitter has no significant affect - random jitter gives a tiny insignificant fixed noise. Its why I don't go endlessly on about femto clocks as the DAC is innately jitter insensitive. There are many more problems with noise shaping, as it is a very complex subject, but this will give you a flavour of the issues involved.
Is the sound of the hugo due to the filter or due to filter/dac combination?
The sound of Hugo is down to lots of things, but of course the primary problem that Hugo addresses is the time domain one. That's where we are converting the sampled data into the original un-sampled continuous analogue waveform - the original signal at the ADC sampling point. Now we are trying to re-create the original un-sampled waveform - re-creating all the missing bits of data from one sample to the next one. Now the theory is very straightforward - if you use an infinite tap length FIR filter with a sinc impulse response you will absolutely and perfectly reconstruct the bandwidth limited signal - if its perfectly bandwidth limited to below 22.05 kHz it will not matter if you sample at 22 uS or 22 femtoS it will make no difference to the output - if you use an infinite tap length FIR filter. Now of course, we can't have infinite tap lengths filters, we have to make do with something very limited.
The question is, what level of time domain accuracy do we need where improving it makes no difference to the sound quality? That's where lots of careful listening tests comes in, as nobody knows. And its where I have been spending a lot of time over the last 18 months working on project xxxx - and I have learnt a lot (and I still have more things to discover, I am sure that I have not gotten to the bottom of the time domain accuracy barrel). What is clear to me, is that the ear/brain is amazingly sensitive to tiny time domain errors - there does not seem to be a level which one can say is insignificant. This is one of the really weird and interesting things about correlating what one hears with real signal errors - the other really odd issue being the perception of sound-stage depth - this can be upset by seemingly impossibly small errors.
This is where I find the "DAC bit perfect" concept - like a cheap politicians sound byte - ridiculous. The job of a DAC is to reproduce the continuous waveform at the ADC sampler -
NOT to bit perfectly reproduce the sampled data with all the sampling time domain errors perfectly intact.
If you were to use this filter with a conventional resistor ladder DAC would it work?
The answer to this is yes, but not as well as Pulse Array - the 16e DAC can reproduce 50 MHz sine wave albeit with 3% THD and noise! The problem with R2R is that the OP can't switch fast enough, as there are a lot of switches involved in the R2R ladder, so in practice you can't run them above 16 FS - but I can run mine at 2048 FS so the digital domain is much closer to the original un-sampled analogue waveform. There are lots of other problems with R2R - noise floor modulation, code dependent glitch energy, high distortion at small signal levels, and moderate distortion at large signal levels.
I hope I have not confused things too much - but we are dealing with a very complex subject, and something which, after more than 30 years of intense work, I am still learning new things. Things are very complex when you dive into it, and the ear/brain is a remarkably sophisticated device - the illusion of listening to real sounds is a truly amazing brain construct, and its something we know very little about. But at the end of the day, the engineering that goes into Hugo does not matter, its the musicality that counts, so keep on enjoying music!
Rob
