This thread will server as a “blog” of sorts on my journey to planning and building a discrete headphone amplifier.
Concept
There are various amplifier classes, some suited to audio applications, others not so much.
Design Corner Stones:
(These should be nailed down at some point)
-
Switchable Class AB/H - At the core of a Class H amp design resides a Class AB amp. As the market is flooded with discrete or integrated AB headphone amps. I do not (= can not) compete with those.
-
Fully Discrete - As this is a learning project first, potential kit/product second, I want to be able to stick a probe on every leg and observer what is happening.
-
No direct Grid feed - May sound odd, but an off the shelf wall adapter makes certification easier.
-
Do not trust the Power - Surges, EMI, switching noise, etc. will be filtered out before the internal linear supply gets to do its job. The aim is to build something that does not care if there is a pair of car batteries or a thousand dollar precision source meter supplying power.
Hurdles
(There are always more of these, in decending order of importance)
-
Certification - When I sell this as a kit with plans, I only need to declare CE-conformity and I am good. Pre-built units would need further certifications like “TÜV” or FCC (I could use some expert knowledge on how exactly FCC conformity works).
-
The PSU - I need 4 rails for this to work correctly in Class H mode. As I painted myself into a corner with the external wall adapter criteria, I will have to create a so called virtual ground before the main PSU can generate the 12V, 5V, -5V and -12V rail. These voltages are arbitrary.
This virtual ground either result in a device that always has ground loop issues or a device that never has ground loop issues. I lack the experience to tell which one of the two.
Just one for the “meaningless feature”-list is that I will either use a “precision” zener diode or reference voltage source. -
Input Impedance - More of a soft goal, but I want to present an input impedance of at least 10k Ohm.
– The cheap route would be a 10k resistor before the main volume pot.
– Another way would be to have an Op-Amp with unity gain sense from a 50k resistance. Kind of breaks my “fully discrete”-goal. I could get an integrated AB-amp chip and run it into a voltage divider to get it to unity gain.
– [Non-Option] Let the user deal with “whatever” impedance. -
Mode Switching - To reconfigure from AB to H mode, I need to cut out the additional transistors and put the Diodes directly in front of the inner transistors. As I do not want the user to flip multiple switches or place jumpers, I need bi-stable relays. In doing so, I need to prevent reverse-biasing transistors (bad!) or shorting supply rails (not good, but LDOs current-limit anyway)
Rough sketch of Class H to AB switch:
I can not just disconnect the high voltage supply rails because Class B would be BAD (then again, being the worlds only Class AB/B/H headphone amp might be cool?)
One upside of having all these relays is that I could present the user with a nice 4-position switch to select B, AB (low voltage), AB (high voltage) and H. -
Volume Pot - I put several minutes of thought (
) into this, looked at a grand total of 4 spec sheets and then decided I don’t want stepped attenuators, digi-pots or exotic solutions. Current idea is to have a pair of multi-turn pots connected via timing belt (I could not find one with two channels). The idea here is to get out of the channel imbalance while there is nothing audible going on while retaining the fine controll of an analog pot.
Thoughts
Cost should be manageable. I have not compiled a BOM, I expect less than 200€ though.
Size wise, it should fit on a pair of “normal” sized proto boards, one for the PSU, one for the Amp itself. That is okay for prototyping, for kits or products, not so much. Form factor could either be “box” (think Vioelectric V280) or “flat unit” (think RebelAmp).
Why?
Because I am a tinkerer. And because all the people using speaker amps for power-demanding headphones makes me slightly irritated.
When?
Dunno, mid next year?