Tube DIY Asylum: Mo' Thoughts on HF noise by Lynn Olson

Some of this is influenced by my nine years of working at Tektronix, five of them in the Spectrum Analyzer division. I worked on the 492 and 496 series analyzers, which were portable mil-spec instruments that retained full specs (0 to 1.8GHz with -80dB noise floor) while being subjected to heat, shock, and vibration. It was also highly resistant to incident RF, enough so that it could be used next to a full-power military radar dish while retaining a clean display. In terms of internal noise rejection, the specification required a completely clean display (no spikes of interference) despite using a switching supply, an internal fan, digital storage display, microprocessor control, and a triple-conversion IF system. It was designed by a team of forty engineers, five of them microwave specialists.

This is a bit different than the average hifi company with three self-taught engineers and two to five marketers in the front office (which is a description more like Audionics, the firm I left before working for Tek). Consumer stuff, even $100,000 so-called high-end equipment, is on much cruder level of design and build quality than genuine aerospace grade equipment. In fact, you can buy real mil-spec for less than "high-end" gear; a brand-new 492 analyzer cost US$45,000 shipped directly from Beaverton. So I don't take the high-end market very seriously - it's mostly hype, image, and four-color ads that cost $5,000 each.

One thing the Tek engineers taught me was what they called "designing from first principles." This basically means figuring out what the heck you want to do before you start, then thoroughly analyzing the behaviour of all the parts you want to use. This is very different than recycling minor variations of old textbook designs and using old "rule of thumb" approximations. It means you scope out the problem, and find out the limitations of the technology you have available. In rare cases you might pioneer new technology, but this approach is risky and endangers the whole project by relying on "blue-sky" technology that may have hidden problems that might emerge late in the project.

Going back to the "first principles" approach, you don't sweep problems under the rug by making assumptions that such or thus isn't audible. That's fine for Dolby Digital or MP3, but isn't good enough for high-quality audio. Distortion matters; noise matters, and a good first approximation is to use the Fletcher-Munson curves to frequency-weight the audibility of distortion. Correspondingly, low-frequency distortion (20 to 120Hz) is 20-30dB less audible than distortion from 1 to 8kHz.

This means sum-and-difference IM distortion sidebands may well audible below the thermal noise floor; the BBC and other organizations have found that tones can be audible as much as 10dB or more below a random noise floor. A few researchers claim that tones should be at least 20 to 30dB below the noise floor in order to portray subjectively high-quality depth and ambient-space impression.

Since we already know that distortion rises quite rapidly with frequency in audio equipment (mostly due to capacitive loads), that tells us we want to avoid stressing an audio playback chain with unwanted spurious HF information. What starts as an innocent-looking 0.1% distortion at 1kHz can rise to a much less innocent 5% at 100kHz.

This means that 88.2kHz of unfiltered hash from the CD player can crossmodulate with 100kHz from a switching supply, creating sum-and-difference IM products at 11.8kHz and 188.2kHz. Note these are the IM products from only the 2nd-harmonic; each harmonic-distortion term generates its own set of different sum-and-difference tones. So in practice, there are lots more than one sidetone that appears in the audio band.

Worse, the CD player and switching supply are asynchronous, so the unwanted sidetone drifts up and down in frequency. If we're talking about a generic switching supply, the switching frequency will drift under load, so the sidetone is then program-modulated (much worse). Kind of like an off-key flute player in the next room that has no sense of music at all, but insists on playing along with the orchestra.

The extremely low intrinsic distortion of DHT's (lower than any other active device) is a blessing and curse. You can hear the music with astonishing clarity, but you also hear power-supply crud too. For example, wideband "hash" from solid-state diodes lends a subtle "grey" coloration to the sound, which most of us take for granted until we hear it's absence, which then sounds like music in a real room with real instruments. Hearing the breathing of the performers - even on bad recordings - tells you you are getting close.