Yggdrasil: The back-saga, Part 1
Here I was, all dedicated to finding out why the CD player I bought in Japan sounded so terrifying. There was at that time, only one problem – I didn't know s..t about digital audio as applied to hi-fi. OK, OK, I had worked with the first, exploring for oil, digital audio recorders in Peru; they were 16 channels of 8 bit log encoding, 5 Hz sample rate machines. They were for measuring ground motion as a function of increasing distance from a subterranean explosion. As far as I knew, digital audio worked great for finding pockets of oil for which to drill. So it would seem to be reasonable that it oughtta sound right for hi-fi, no???
So I studied and researched what I could in a pre-internet era. There were a few things that I accepted as creed, chiefly that the more technology changes, the more the requirements stay the same; this explains the syndrome where new doo-dads promising ever better performance often disappoint; the technology takes precedence over the application. I learned that the vacuum tube, the transistor, the speaker, and the cartridge, was either invented or perfected by Bell Telephone Labs. The way that it used to be back in the day was that Bell Telephone Labs had a monopoly on all home phone service in the US. In today's dollars, a month's worth of phone service where you had a calling radius of 20 miles or so cost about $100 or so. What that did was subsidize one hell of a think tank that was dedicated to developing electronics goodies; computers, TVs, radios, etc., etc. They made it possible, among many other nifty hardware apps, to hear announcers at baseball games, not only in the ballpark but at home!
Turns out they also developed and propounded information theory and digital audio well before there was hardware to implement same: papers date from 1912 to 1916. This even before the US got sucked into World War 1! But I get ahead of myself. After all, you have to crawl before you can walk. I had to look at current CD players. There had to some things one could do to make digital audio palatable.
It turns out that the D/A converter chips are the most expensive chips in the whole CD player. The first CD players in Japan cost 380,000 yen, or north of a kilobuck! A great way to save money was to use only one DAC chip function for both channels – one stereo channel converted and then the other; no kiddin'! No way for the two channels to ever be in proper phase. That had to be it! So I bought another DAC, duplicated everything else, and fed the DACs (now 2) at the same time to fix the coherence problem. It got coherent! There was an incremental change. I was excited! Mike came by, listened a bit and curled his lip. I put the turntable back on – he was right.
Next I got rid of sample and hold amps and replaced them with brick walled amps. Another improvement! Mike came by again, curled his lip again. Paul showed up, listened for about 10 seconds. He then opinined to no one in particular: ”The corn kernels are gone, but it's still a turd.” Mike, his sensibilities temporarily abandoned, nodded approvingly.
The solution had to be simple, I told myself. By this time, 2nd generation CD players were out. I settled on a Toshiba Model with 1 dac chip per channel, did my usual mods and hooked up my scope to the digital end of the dac. I thought that those were just awful looking fuzzy edges on the digital waveforms. No one was calling it jitter yet, but it occurred to me that that would convert to some really random wow and flutter (a forgotten analog turntable and tape recorder spec that really is not obsolete in the digital audio era).
So I reclocked the whole deal from the dac chip back with crystals and plls. The digital waveforms were far less fuzzy. I hooked it up yet again. I listened for two or three evenings before I convened the cast of usual judges. Mike listened quietly, while Paul picked his nose. Both were quiet, requesting more music, inscrutable faces. After an eternity, Mike exclaimed: “There's actually an image”. Paul grunted in agreement. “Violins still kinda suck”, Paul said, “but its kinda ballsy for rock”.
That was fine with me. Progress, I thought. I began to recommend and modify these players for an ever widening group of audiophiles. One of them, Dick Olsher, wrote for Stereophile and gave me some kind words. Everyone basically said the same thing: “Not bad for digital”
Another one of the early adopters was a math professor at the University of Iowa named John Lediaev. Classical music geek and true TWAD (Tree Worshiping Analog Druid). He got interested in the whole digital audio sojourn and gave great input.
Meanwhile, Phillips came out with a multirate 4x chipset and CD player! (Multirate=different input and output digital frequencies, in this case 4x up or oversampling.) This meant that the signal rate to the DAC was 176.4KHz instead of 44.1KHz, allowing a much gentler analog filter than the brickwall filters required before which by comparison rang like old telephones. So I rushed to buy one of these new boxes, hooked it up and listened in amazement at the total lack of consistency in any property at all: image, tonal balance, etc. The only saving grace was a less harsh/edgy top end – logical for the much softer filter. The oversampling approach used a Phillips provided dsp chip. In those days, it was called a digital filter chip.
Reverse engineering of the Phillips digital filter chip, (as well as virtually all other subsequent ones by other manufacturers – up to this day) all utilize a frequency domain (read flat) algorithm which has no mathematical solution, only a succession of approximations. (read close – horseshoes). Worse yet, it does not keep the original samples – it throws them away and approximates new ones. Ugh.
I wondered why all of these filters used this method of derivation. In a flash of brilliance reserved for the obvious, it came to me that this algorithm is efficient (cheap) and gives good measurement (very flat). There were also a variety of toolkits and developmental software available to buy devoted to this algorithm only. This simplified things - making it relatively easy to for anyone to develop such a filter. There was no need to hire and babysit socially handicapped genius savants who needed cages as well as food and water pushed under the door from time to time in order to develop a proper filter.
I decided to consult my old Bell Telephone Labs 70 year old at the time references. Wow, not only did they discuss the common frequency domain one, but a time domain one as well! Now here we go! Well, the problem was the time domain one was that the frequency response curve looked like an anthill. (It turned out later that a competitor actually used this very algorithm)
Out of this knowledge two new conclusions arose. The first was that oversampling digital filters were important - to keep the decoded audio from ultrasonically cleaning windows and looseingn tooth fillings. The second was that whatever audio information remained in the original audio samples could not be lost.
I called John Lediaev back and told him we had to combine time and frequency domain optimization for digital filters – he told me it couldn't be done because it required a divide by zero and by the way why did I want to do it anyway. So I told him I wanted to beat the sound of analog with digital audio. Instantly he said “There's no way – it can't be done”
A stubborn kernel of determination blew up in me like an airbag. “**** him”, I thought, “I'll show him” One of the manifestations of my audio insanity is that if anybody ever tells me I can't do something – I will move heaven and earth to get it done. But there was a problem: intuitively I knew I couldn't get it done without the combination of the two algorithms with the original samples preserved. If there were anyone on the planet sufficiently motivated and capable, it was him. A bonus was I really like and admired the guy and wanted to work with him.
Soooooooooooooo I told him, I bet you can't screwin' combine the two optimizations and keep the original samples. Silence.................... More silence. I had him!! He finally said “Give me some time”
But I was still overburdened; I knew a DSP engine of some kind had to be built and programmed. This was a big deal in the mid 1980s. It was a lot of research and hardware. So a week or so later I was reading all about TI DSP processors when I heard a noise that sounded like a combination of a twenty year old power lawnmower and a coffee can full of nuts and bolts slowly rolling down a steel staircase. The turquoise 1964 Toyota Corona out back with white smoke coming out of its exhaust and black smoke coming out from under the hood confirmed it. There was only one person who I knew that drove cars that were so “****ed-up” that even cops, reluctant to deal with crazies, balked at stopping them. I used to wonder how he got so broke, but later came to realize he was just cheap. I mean, this car was such a loser that you couldn't even give it to a charity, even if you hauled it to them. In any event, out stepped a grinning Dave and I knew the team was complete.
Part 2 should be next week, according to Yggy.