Superlux HD668B SE classA amp
My Superlux HD668Bs are the best value purchase I've ever made in headphone kit, far and away the best bang for the buck at their very modest price-point. If any pair of cans is worth designing a specific amp for, they are most certainly on the short-list. The amp I designed specifically for my AT MSR-7s doesn't have enough ooomph for these as the impedance is appreciably higher and the efficiency lower.
So here is my attempt at a dedicated HD668B amp with no corners cut in SQ while still being cheaper to buy the parts for than the cans themselves. I've designed it to run from a single lithium ion cell - a Panasonic 18650 would work fine and give around 60hrs of listening time. Its a single-ended classA topology driven by a single current-source loaded MOSFET. And naturally enough it sounds like no amplifier at all - a straight wire without gain, seeing as its just a buffer.
In actual fact it certainly does have gain, but only current gain. And no shortage of that given its very high impedance input made possible by use of a MOSFET. I've made bipolar versions of a very similar circuit in the past - they've been limited by the transistor needing appreciable base current. Using a FET, gate current is immeasurably small at DC and LF.
With single-ended classA amps, its important to know the load impedance, which is why I've designed this to drive the Superluxes, which are 56ohm. Other types whose impedance is similar can be driven - going lower impedance though would necessitate circuit mods to enhance the available drive current.
As I've already mentioned, the centrepiece of this amp is the MOSFET, one I've never seen used in audio. Its design application is as a high-sided power switch, not anything in linear mode. However what struck me as odd is that even though its designed to be used as a switch, there's a transconductance vs ID plot, one of the best I've seen anywhere in the realm of FETs. Its impeccable credentials for gm seemed to me to be crying out for an audio application... I'll not be revealing the part number just yet - I don't want the price to suddenly shoot through the roof (and fakes created) because its become a fashionable device
The part showing in the LTSpice schematic is a fairly close substitute though not one I've tried.
Update 1 - the current source limits the maximum positive output swing to around 0.8V below the PSU rail, which is a loss of over 20% with a 3.6V rail. I'm investigating use of an opamp-controlled CCS with a much lower drop-out voltage around 200mV. If you'd like to see a schematic leave a note in the comments.
Update 2 - I've uploaded a pic of the 4th prototype, the second variant to use opamps for the CCSs. Most of the passive components are on the reverse side - the opamps are SOT-23-5 packaged.
Update 3 - I have now built four of the 2nd generation variant, the next step is trying out paralleling a few stages to get more power.... Here's the first prototype 5 slice (600mW) amp taking shape. Idle current is 125mA per channel @20V supply. We shall have to see if an input buffer is called for to isolate all that parasitic gate capacitance....
Update 4- some good news - so far no sign of oscillation and the slices share current nicely. Which brings about some advantage in that the gm of FETs per unit drain current always decreases with increased current. So the way to get highest gm overall is by running each FET at a lower current and paralleling them. The downside of course is increased parasitic capacitances. I have been using FAN4174 opamps for the CCSs, these are cheap but not the cheapest so I'm looking out for even cheaper CMOS opamps than these. MCP6001 is a possibility.
Update 5 - another way to increase the output power is by using some beefier MOSFETs. I've uploaded two pics of my 'hot buffer' which is using TO220 MOSFETs running at 500mA idle current. I predict that the output impedance is going to be of the order of 0.5ohm which is pretty good for a zero GNFB amp
Meaning these FETs have a gm of at least 2.
So here is my attempt at a dedicated HD668B amp with no corners cut in SQ while still being cheaper to buy the parts for than the cans themselves. I've designed it to run from a single lithium ion cell - a Panasonic 18650 would work fine and give around 60hrs of listening time. Its a single-ended classA topology driven by a single current-source loaded MOSFET. And naturally enough it sounds like no amplifier at all - a straight wire without gain, seeing as its just a buffer.
In actual fact it certainly does have gain, but only current gain. And no shortage of that given its very high impedance input made possible by use of a MOSFET. I've made bipolar versions of a very similar circuit in the past - they've been limited by the transistor needing appreciable base current. Using a FET, gate current is immeasurably small at DC and LF.
With single-ended classA amps, its important to know the load impedance, which is why I've designed this to drive the Superluxes, which are 56ohm. Other types whose impedance is similar can be driven - going lower impedance though would necessitate circuit mods to enhance the available drive current.
As I've already mentioned, the centrepiece of this amp is the MOSFET, one I've never seen used in audio. Its design application is as a high-sided power switch, not anything in linear mode. However what struck me as odd is that even though its designed to be used as a switch, there's a transconductance vs ID plot, one of the best I've seen anywhere in the realm of FETs. Its impeccable credentials for gm seemed to me to be crying out for an audio application... I'll not be revealing the part number just yet - I don't want the price to suddenly shoot through the roof (and fakes created) because its become a fashionable device
The part showing in the LTSpice schematic is a fairly close substitute though not one I've tried.Update 1 - the current source limits the maximum positive output swing to around 0.8V below the PSU rail, which is a loss of over 20% with a 3.6V rail. I'm investigating use of an opamp-controlled CCS with a much lower drop-out voltage around 200mV. If you'd like to see a schematic leave a note in the comments.
Update 2 - I've uploaded a pic of the 4th prototype, the second variant to use opamps for the CCSs. Most of the passive components are on the reverse side - the opamps are SOT-23-5 packaged.
Update 3 - I have now built four of the 2nd generation variant, the next step is trying out paralleling a few stages to get more power.... Here's the first prototype 5 slice (600mW) amp taking shape. Idle current is 125mA per channel @20V supply. We shall have to see if an input buffer is called for to isolate all that parasitic gate capacitance....
Update 4- some good news - so far no sign of oscillation and the slices share current nicely. Which brings about some advantage in that the gm of FETs per unit drain current always decreases with increased current. So the way to get highest gm overall is by running each FET at a lower current and paralleling them. The downside of course is increased parasitic capacitances. I have been using FAN4174 opamps for the CCSs, these are cheap but not the cheapest so I'm looking out for even cheaper CMOS opamps than these. MCP6001 is a possibility.
Update 5 - another way to increase the output power is by using some beefier MOSFETs. I've uploaded two pics of my 'hot buffer' which is using TO220 MOSFETs running at 500mA idle current. I predict that the output impedance is going to be of the order of 0.5ohm which is pretty good for a zero GNFB amp
Meaning these FETs have a gm of at least 2.Total Comments 15
Comments
-
Must be be a pretty big die fet to get that kinda gm at Low currents. Which makes me wonder if the input capacitances were a problem ?Posted 23rd May 2017 at 04:20 PM by kasey197
-
Good point. Here are the figures for its capacitances 290pF, 44pF, 32pF (input, output, feedback respectively).
Note that for a source follower with relatively high gm (at a bias of 25mA its around 0.5S) the input capacitance is bootstrapped due to the source following the gate quite closely.
In order to get some idea of the capacitance seen by the driving source I have checked the input current in LTSpice - its about 12uA RMS @ 10kHz, 1VRMS in. The Si5515 capacitances are slightly higher than the FET I'm using (455,70,54) so I'd expect slightly lower for my physical prototype. The 12uA figure seems to suggest we're under 200pF input capacitance.Posted 24th May 2017 at 12:54 AM by abraxalito
Updated 25th May 2017 at 04:25 AM by abraxalito -
Thank you for another interesting entry Richard. I guess with your experience you are able to tell what would cut corners into SQ, and what wouldnīt. For my benefit as well! I might stop overcomplicating some circuits and try to find simpler solutions like this one.Posted 25th May 2017 at 03:09 AM by Alexandre
-
Thanks for your encouraging comments Alex. I've found that a small improvement can be made by increasing the coupling caps values (33nF input, 470uF output). So I'm going for 100nF input, 1500uF output on the next rev so as not to cut corners
Running this in LTSpice seems to look not so good at full output, but then I remembered music spends almost no time at all at maximum level, dropping the level by 6dB and all looks very respectable....Posted 25th May 2017 at 03:49 AM by abraxalito
-
These are quite interesting curves -but the si5515 is not something accessible to most - its packaging is a bit hard to work with diy...
I'm in the midst of a project that uses p channel mosfet as output so if you flick me a pm with the part number I'd be happy to give it a whirl. I won't broadcast it of course :0Posted 25th May 2017 at 03:06 PM by kasey197
Updated 25th May 2017 at 03:09 PM by kasey197 -
I'd want to be sure that the part I'm using is suitable for your application. What power dissipation do you need?Posted 26th May 2017 at 12:43 AM by abraxalito
-
Around a watt or two is about what I need ...
sperately don't you think running yours at such low Vds is inviting distortion ? Or does it measure ok ?Posted 26th May 2017 at 02:51 PM by kasey197
-
A watt or two is going to be too much for this little chappie, he's only a slightly under-sized SOT-23. Sounds like you'd need SOT-223 as a minimum.
I haven't done any measurements yet but what I've done in sim is subtract output from input and looked at the residual. I'm in general not confident that sims return accurate enough THD and anyway consider that number mostly irrelevant to gauging audio quality. Its a good check if there are any design mistakes though.
Low Vds is inevitable with such low voltage rails as from a single cell.Posted 27th May 2017 at 01:26 AM by abraxalito
-
That's understood and agreed on the device dissipation. Looking at the chart you posted 1amp at 3vds I thought this was a little bigger :00
I would approach this design only very slightly differently - with higher rail voltages from a dc-dc convertor after the 18650.
You know your stuff though so good luck with this I'll be watching with interestPosted 27th May 2017 at 02:56 AM by kasey197
-
I used a charge pump converter on a classAB headphone buffer (elsewhere on my blog) to get 7.2V - that one had much lower swing due to the use of cascoded output devices.
I searched for some higher dissipation FETs which might be suitable for your application, drew a blank as none in SOT-223 came close for gm to this one I'm using. I figure to go up in output power its probably possible to parallel these babies. Since the threshold voltages are bound to vary rather a lot, paralleling them by AC coupling is probably going to work best. If I had some headphones that needed more grunt I'd even build a prototype, but right now its only my 600ohm DT880s that can't be driven by this. Those Beyers need much higher swings.Posted 27th May 2017 at 03:36 AM by abraxalito
-
Thanks for looking for that fet in sot-223 ;0 let me know if you think I can get away paralleling maybe three of the smaller ones?Posted 28th May 2017 at 11:23 AM by kasey197
-
My first four prototypes I built with the FET in a UFM (2mm long) package which is rated at 500mW with about 600mm^2 of copper on the pads. Subsequently I've found the same die can be gotten in a true SOT-23 where the dissipation rating goes up to 1W with the same amount of copper. So if you need 2W of dissipation (which would correspond to 1W of classA output roughly) then three transistors paralleled should be able to manage comfortably. The 1W rating is only valid at 25oC ambient which would be very hard to achieve inside a case.
The opamp-based current source is using the same P-channel MOSFET so you'd need a total of 12 transistors for a stereo amp. Fortunately they are extremely cheapPosted 28th May 2017 at 01:50 PM by abraxalito
Updated 28th May 2017 at 01:56 PM by abraxalito -
Ok thanks ! that's sounds perfect.... don't keep the suspense going for too much longer tho ;0 let me know and I'll put this into my next parts order
My fave p-ch has been the bsp92 for some time now - time to give it some competition (pdiss notwithstanding) !!Posted 28th May 2017 at 03:10 PM by kasey197
Updated 28th May 2017 at 03:13 PM by kasey197 -
Having looked at the BSP92 DS I can see why you'd be worried about not having enough rail voltage - because of its higher breakdown voltage the on-resistance is considerably higher. Although the gm plot looks good my guess is it was made at a fairly high Vds (probably 10V at least) and when operating with only a couple of volts across the FET the gm is going to be considerably lower.Posted 29th May 2017 at 01:21 AM by abraxalito
-
Richard, today I have tested a few LPF options on a balanced transistor I/V for tda1543, single supply. With no LPF the sound is quite enjoyable off axis: I might prefer no filter if I am away from the speakers (cooking for example).
At the listening position some filtering is needed to make the sound more natural (my speakers are quite forward). Right now I have two Vishay KP 2.2nF 250V connected in series (acts like a 1.1nF therefore). Connected across the phases (between I/V resistors) and the midpoint I can connect to gnd, vcc or leave disconnected. Once again I find the sound more enjoyable and easier to follow with the midpoint disconnected. Itīs that old story: donīt filter "to ground" with these chips!
I will place a jumper between the two caps and keep casually comparing filtered vs non filtered.
Another idea is a common mode choke before the filter cap, then take the output from the cap. I have a ferrite toroid that might be suitable (1 inch diameter). No idea about the parameters of it though... What do you think Richard?
BTW my amp is completely reworked, its a single supply bridged THS4131 plus two LM3875, all inverting, all chips receive the same 32V regulated supply (zener -> 2K2 -> 1200uF -> triple EF), and the raw supply is 40V 20mF capacitance. Sounds very good.Posted 11th June 2017 at 12:55 AM by Alexandre
Updated 11th June 2017 at 01:01 AM by Alexandre
