Neurochrome HP-1

At ±24 V, life is much easier indeed. It's just that there aren't really any technical advantages of a linear supply. The switchers are less expensive too. There may be a marketing advantage of the linear supply. Some fear switchers - and for a good reason. There are horrible switching implementations out there. Mine is not one of them. I hope my measurements will put any doubt about the capabilities of switching supplies when in the right hands to bed once and for all. You can also see my measurements of the MiniDSP 4x10HD. They use an external switching power brick and a couple of switching power regulators inside the chassis. No issues there either.

Tube switching supply. Oh yeah... That's a back burner project of mine. I offer switching filament regulators already. Mine switch at 500 kHz. See the Universal Filament Regulator.

Tom

 

Oh, I bet. Very interesting bio too.

 

It was brought up earlier, but I'll ask it again now: What are your thoughts on driving headphone/speaker loads in balanced vs single ended mode.

 

At low volumes do you experience any channel imbalance?

 

That is a relevant question with that miniature pot.

 

I can not perceive any difference between the two channels on a listening test.

I can measure a difference. From about -50 dB and up on the volume control, the channel tracking is within ±0.1-0.2 dB on the prototype unit. Near max volume the two channels track perfectly (better than ±0.01 dB). At volume settings below about -50 dB the channel tracking does get a little worse. It's a potentiometer, not an attenuator after all. The worst spot was around -75 dB where the two channels were a few dB apart. -75 dB is just off the minimum volume setting.

Tom

 

I'm doing my chassis in RAL 5010 Genetian Blue. Behold, gentlemen:

 

Definitely. I did look at the specs before choosing that pot.

The RK097 is specified to ±3 dB (-40 -> 0 dB). Mine measures way better than that. The RK271 "Blue Velvet" pot is ±2 dB -60 - > 0 dB and ±3 dB -70 -> -60 dB. The RK271 is a slightly better pot, though I'm not sure it's meaningfully better. Unlike the RK271, the RK097 is commonly available in large quantities in 10 kΩ, audio taper. The RK271 is often only available in 100 kΩ (at least at Mouser). The 10 kΩ pot is already the dominant noise source in the HP-1, so I'm definitely not going with a 100 kΩ pot. I'd use a 2 kΩ pot if anybody stocked it.

There's another advantage of the RK097: The pins have little leaf springs on them so the pot snaps in place. That's very handy from an electromechanical standpoint as pots tend to be torqued loose from the PCB with years of use. That's much less likely to happen with the RK097 than with the RK271.

Tom

 

Thanks for getting back about that. Yeah there isn't a wide range of choices for pots rated below 50k in that size. I imagine manipulating the gain would suffice for avoiding channel imbalance.

Out of curiosity, would this be a viable fit into the board?
http://www.partsconnexion.com/tkd_73334.html

 

According to the spec sheet, the pin spacing is different so it won't fit in the board. It's also not a better pot than the RK097, so you're not gaining anything except a much lighter wallet.

I have not had any issues with channel tracking in my use of the HP-1. About 20-25 people listened to it using a wide array of headphones and IEMs at the Calgary Head-Fi meet two weeks ago. People praised the amp for its neutrality. If there had been issues with the channel tracking, I'm sure someone would have said something.
Also note that I designed the amp to have relatively low gain (6/12/20 dB selectable by switch). I recommend running at the lowest gain setting that gives you satisfactory volume. This means you'll tend to be fairly high up on the volume control (certainly past 10 o'clock), thus, set yourself up for the best channel tracking from the start.

If you go the DIY route, you can always connect your favourite pet pot or attenuator to the side of the board using very short wires. Just be careful as that's the most sensitive node in the circuit.

Tom

 

I see no advantage of differential drive of headphones. In theory, you should get cancellation of the even order harmonics (assuming the two channels of the differential circuit are perfectly matched). That'll lower THD, which would look good on the spec sheet, but leaves the odd order harmonics that tend to be favoured the least in listening tests. I'd also point out that I'm measuring -135 dB THD already. I'd be a bigger proponent of the harmonic cancellation if my circuit actually created any harmonics worth cancelling. Also note that most cancellation circuits look great on paper, but are mediocre if not downright detrimental to performance in reality. At least that's been my experience designing precision circuits for a living for over a decade.

I do see some advantage of using the 4-pin XLR over the 3-pin phone connector, however. That little bit of shared ground impedance in the 1/4" phone jack actually degrades the channel separation from 115 dB to 90ish dB. That's physics for you. There's no magic voodoo here. It's all Ohm's Law.

Tom

 

Hey, Tom! Welcome to SBAF! I've been closely watching your activities on DIYA and got a Modulus kit from you.

The HP-1 is a nice design. I especially like that you've added various safety features. Sadly it's a rare sight in DIY, especially in headphone circuits.

Some questions:

I see that you've implemented DC servo circuits in an IC based amp. Was DC a problem during everyday operation? I see that they might be necessary with discrete circuits where more thermal-borne offset effects are present. IC amps have never given me any trouble. And I've even managed to dial in discrete class-A circuits with unmatched components to <5mV.

Given your experience with IC based circuits the component choice for your non-tube circuits isn't a surprise for me. My Modulus amp was able to go toe to toe with just about any discrete amp, so I've got no problem with it. What's your take on discrete vs. integrated?

 

Do you believe that current measurement suites are sufficient to represent the full range of human hearing?

 

Yeah. It's simply due diligence. It also makes me sleep better at night.

In a DC coupled amp, a DC servo is absolutely necessary. If your source has 5-10 mV offset (maybe it uses an old opamp for the output driver) and you use the HP-1 at the 20 dB gain setting you'll get 50-100 mV on the output. That's way too much for a headphone amp. With the DC servo, you can apply a few volt of DC to the input of the amp before the output of the amp moves away from 0 V.

Some offsets arise from mismatch. Some arise from the base current of BJT inputs. You can get rid of that by using a MOS input, but then you're trading off low noise in exchange for low offset. That's not a good trade in a ultra-high end headphone amp.
I'd also argue that < 5 mV isn't exactly high end for a headphone amp. It's fine for a speaker amp, but for headphones I prefer < 1 mV for sure.

I'd rather not engage in a religious discussion, so I'll word my answer with some care here. IC design has a lot in common with discrete design. The fundamental topologies used are identical in many cases. The main difference is that in IC design, the matching between devices can be stellar (if you make a good layout) but the absolute component tolerance is not as good. Thus, IC circuits depend on matching (among other things) to achieve good performance. In discrete design, the component matching is usually rather poor (compared with the matching you can get on an IC anyway), but the absolute tolerances are stellar. These days ±0.5 % resistors cost the same as ±1 % resistors and ±0.1 % resistors are only about $1 each. So discrete circuits rely on the absolute component values to a greater extent.

The stellar matching and the possibility of having high-impedance nodes on an IC without destroying the signal integrity makes it possible to design an amp with very few gain stages that has stellar linearity and an incredible amount of gain. In discrete circuits, the nodes are long and leaky, so you really want to avoid high-impedance nodes. On top, the devices are slow and have relatively low gain. Thus, you end up with many gain stages to get even above 60 dB of loop gain. Compensation becomes a challenge.

I have yet to design a discrete amp which outperforms an IC amp.

I'd rather not engage in a religious discussion, so I'll word my answer with some care here. I feel like I'm starting to repeat myself... My Audio Precision APx525 is way more sensitive than the human ear. On top of this, there is an overwhelming body of research supporting the argument that good measurements (plural) correlate well with a good listening experience. So, yes, the "current measurement suites" do provide the data necessary to make informed design and product choices. The only issue I see with the "current measurement suites" is that some individuals lock in on one number rather than seeing all the measurements in context. That's why I emphasized the plural 's' in measurements.

Tom

 

I'm sure there's a possibility. It's just a matter of how much money someone is willing to throw at a stencil. Personally, I don't have a need for a solder paste stencil. I just use thin (0.5 mm diameter) solder and a 1 mm chisel tip on my soldering iron (METCAL MX-500). A good pair of tweezers is a must as well. I use a titanium pair by Technitool.

Sadly the exciting new parts are all SMD. We'll have to deal with it. The HP-1 mostly uses 0805 and larger parts, which are *huge* by modern SMD standards. There are a few (12-14) tiny diodes in SOD-323 package (aka SC-79 if I recall correctly). They're damn tiny. If you can deal with those, the rest of the board should be no biggie.

Tom