Presonus HP4 modifications

This was my first major modification project, something I’ve been meaning to post about for the last couple of years––it’s morphed over time, I’ve added and removed components as I learned from other projects. The Presonus HP4 is an excellent and powerful headphone amp, and can be found for cheap on Ebay, Reverb, etc. I think I bought this one for about $20 because was missing its power supply or something. Sounds great stock, delivering a whopping 130mW of power to each channel, and there are two designs that I’ve seen: an older version (the one I have), pictured first, which uses six quad MC33079 op amps (DIP package), two for the input and monitor output buffers, and one for each channel (two op amps paralleled per channel)…

Below, the newer version with four quad MC33079 (SMD), two again for the input/output, two for gain stages to headphone amps, and a discrete transistor stage to drive headphone current (I don’t know what the transistors are, so there’s no accounting for noise performance, etc.). Not my picture, taken from this thread:

The MC33079 is fine, old design meant for audio. Like I said, this thing sounds good already, it doesn’t need help… but OBVIOUSLY I had to find out for sure!

Anyway, I was inspired by this thread, where someone modded the newer version.

Stock capacitors are all cheap Chang caps: 47uF/35V bipolar at the input and monitor output; 1000uF/25V and 47uF/25V in the power supply; and (at least in my older version) local bypass caps in the headphone amp section, 10uF/25V on the power rails (+/- 15V!) and a 470uF/25V cap on each channel that seems to decouple headphone ground from main amplifier ground.

Replaced the input and output bipolars with 47uF Panasonic SU-A, bypassed on the other side of the board with WIMA MKP 0.47uF (0.1uF would be fine). Because the original chips were DIP, I took the opportunity to design and order some custom dual soic to DIP adapters (which I believe anyone can order as such from Aisler), replacing input buffer with OPA1612. I could have used OPA1642 as well, and perhaps it might have been preferred; there is some justification for a FET-input op amp like this in the input buffer position since the input impedance is 10kohms. This chart from TI demonstrates why, having to do with voltage noise versus current noise. Using OPA1612 was maybe an unnecessary expense, but it sounds great all the same.

I soldered four 100uF/16V Nichicon R7 organic polymer caps to either side of the 0.1uF ceramic bypass caps already installed around the input/output buffers, just to offer some more isolation from the power supply. Note the orientation. These caps have extremely low inherent resistance (ESR) and can deliver current very quickly to the op amps when needed.

In the headphone stage I used my soic/DIP adapters again, installing two OPA1688 chips per channel. A relatively new design from 2015, the 1688 is designed for headphones (esp. mobile applications like portable DACs, hi-end smartphone amps), putting out 75mA of current per channel, meaning paralleled as they are here each headphone channel is getting 150mA. Excellent specs overall, very low THD, etc. You could use other more exotic op amps here with crazy bandwidth and warp speed slew rates, but honestly for headphones there isn’t any need, and in fact there can be significantly diminishing (or negated) returns if you put the wrong op amp in the wrong circuit. This offers a classic and realistic explanation of all this stuff. I think the OPA1688 sounds great and plenty transparent enough for reference use: RME’s ADI-2 Pro uses six of these in each headphone amp, three in parallel per channel.

More power supply decoupling on each channel, originally 10uF/25V, now 100uF/16V Nichicon R7 which barely fit (also added 0.1uF ceramics on underside of board––Presonus omitted this in their design, which is strange, since this is standard practice). Ground decoupling (big cap in center) is Panasonic FC, 470uF/25V––probably could be anything, doesn’t really touch anything other than ground, essentially isolating amp ground from headphone ground. Also pulled original resistors on either side of op amps, originally 120 ohms each (the old standard for headphone outputs), and replaced with 1 ohm resistors. Some resistance is necessary to isolate op amps from capacitive load of headphone cable to avoid oscillation, plus as short circuit protection, but lowering output impedance will make these amps more usable for low impedance drivers like IECs and earbuds. General guideline is to have 8-10x lower output impedance (amplifier) than input impedance (headphone) to avoid any wonky distortions in frequency response––in other words, amp with 4 ohm output impedance shouldn’t drive any lower than 30 ohm headphones, 38 ohm amp no less than 300 ohm headphones, etc. Presonus says HP4’s output impedance is 51 ohms, meaning 400 ohm cans and above will be linear and anything below that will have lumps and bumps in the frequency response. You can read about this here.

Very simple PSU, two regulators, 7815 and 7915, 1000uF/25V and 47uF/25V per rail, used Panasonic FC, but again probably could have used most anything modern and reliable (low ESR, high heat tolerance, e.g. 105ºC)––Nichicon, Panasonic, United Chemi-con, etc. Hot-glued caps together because tall ones were wiggling around a bit.

And finally the underside, showing WIMA MKP bypass caps and ceramic 0.1uF decoupling. 

The whole shebang!

Like I said, this amp is already a great deal, and these modifications really open it up because the basic elements of the design are very good. The sound is clear, clean, vanishingly quiet, incredible separation and clarity of voices, amazing perceived sense of space around and behind my head (listening with HD600), and absolutely no apparent personality of its own. What comes in is what comes out. The OPA1688 sounds spacious, if I were to put a word to it.

(6) 47uF/35V Panasonic SU-A ($0.63/ea)

(6) WIMA MKP 0.47uF (or 0.1uF) ($0.29/ea)

(2) 1000uF/25V Panasonic FC ($0.79/ea)

(2) 47uF/25V Panasonic FC ($0.29/ea)

(4) 470uF/25V Panasonic FC ($0.60/ea)

(12) 100uF/16V Nichicon R7 (or other low ESR organic polymer) ($0.72/ea)

(8) 0.1uF/100V ceramic capacitor ($0.49/ea)

(6) dual SOIC to DIP14 adapter boards (~$10 + shipping, maybe $20 total, so worth getting a bunch in one order)

(4) OPA1612 ($6.41/ea) [or OPA1642 ($2.08/ea)]

(8) OPA1688 ($1.56/ea)

(16) 1 ohm 1% 1/4W metal film resistors ($0.18/ea)

Total = about $80 with shipping and taxes (here’s a Mouser project for anyone who is interested). Again, that’s for this older version with 6 quad op amps. If you modify the newer version it’ll probably be about half that, since you won’t need the adapter boards and fewer capacitors (anyway, I added the four extra for PSU bypass at the input/output buffers, as earlier described). Furthermore, the newer version uses only quad SMD op amps, which limits your options, but ultimately forces you to save money: I’d use OPA1644 for front end and OPA1604 for headphone amp buffers. You can probably get a used HP4 for about $50 (or less), which will probably put you out about $120-130 for the whole project. Conversely, if you consider the cost of a CMOY project, which can easily run up to $50 for lower performance (wonky split 9V rail, for example), $80-90 for Super CMOY (designed around the OPA1688), or $129 for the O2 headphone amp (which puts out 140mA/channel, versus 150mA in my modded HP4), this is a pretty great deal for four extremely powerful headphone amps. Of course, this assumes you are the kind of person who delights in voiding warrantees and has the equipment to do so…

As a final note, I should mention that this design leaves two op amp channels in the input/output buffer section unused. The monitor output of the HP4, a useful feature for using the device as a monitor level control, is ‘impedance balanced’, meaning the ‘ring’ connection on a TRS plug is connected to ground through a 51 ohm resistor––practically speaking, it’s an unbalanced connection. It’s possible I could use that second channel to create an inverted signal and fully balance the output, but I decided to let this rest for now… An idea for someone else!

Advertisement