Upgrading the XuanZu portable headamp
Posted 27th February 2016 at 12:22 AM by abraxalito
Updated 14th March 2016 at 06:23 AM by abraxalito
Updated 14th March 2016 at 06:23 AM by abraxalito
This device is a steal on Taobao, but having had a quick listen last night it could sound clearer. When connected to my smartphone (Meizu MX4 pro) and compared side by side with my 'Buffalito' (not a blind comparison mind) into my SuperLuxes, there were a few notable deficiencies.
First the soundstage air was less apparent. Second there's some sibilance noticeable on voices. And third the background hiss is slightly more apparent and a slight whine comes from the power supply. So I figured - open her up.....
Inside its fairly simple, the more or less standard configuration of a pot, then opamp gain stage then discrete diamond buffer. Which is great because I already have experience with this topology. The power supply is a built in LiIon cell with a boost converter supplying 12V in a single rail and there's a passive rail splitter. The dual opamp is an EL2244, one I've not seen before in such a setup.
First up - the input pot is too low a value at 10k, smartphones really only shine when loaded very lightly (at least mine does) - by which I mean at least 200k. I doubt I'll be able to find a replacement pot above 100k so the first requirement here is to give the unit an input buffer. This could be discrete or another opamp - I might try out a low power JFET opamp to begin with. But I'll need to improve the power supply first so let's examine that.... stay tuned.
If you'd like to buy one and don't have access to Taobao, its available ($50, free shipping) from Aliexpress - https://www.aliexpress.com/item/High-...208.3.2.25Nzu6
Update - I did a quick little experiment to see what the effect on the sound of the 10k loading is. To isolate this I used my 'Buffalito' as a buffer (its input impedance is above 200k) and then routed its output into the XuanZu. The result was that the soundstage bloom came back, but the other two problems (sibilance and extra noise) were still there. So it looks like to get the best ambience retrieval the high input impedance is crucial.
Stage 1 improvement - remove the SMT 'lytic caps (100uF/16V) and reduce the power supply voltage to allow 10V caps to be used. I did this by paralleling a 1.1M resistor over the top of the 270k which sets the output of the 5pin boost regulator. Previously the supply was about 12V, now it measures 9.83V so I can use 10V caps with impunity. Having removed similar caps on other portable devices I was expecting a poor ESR measurement and I wasn't disappointed. About 0.8ohm at 10kHz which I figure is the reason for the supply whine.
Of the four caps originally present, two are supply line splitters and the other two are paralleled as reservoirs for what was a 12V supply. I've now fitted 470uF/10V caps but these have ESR more than 10X lower than those I removed. A quick listen shows the supply line whine is gone - half of one problem fixed. Next up, checking the opamp....
Stage 2 improvement - replacing the opamp. The EL2244 is bipolar and runs about 5mA per side. Since the device's output stage is classAB and I want to minimize the noise generation effects of this I changed the opamp to a JFET type so the feedback resistors can be substantially raised in value. Originally there was 10k feedback, 5k6 to GND. Now I have 470k feedback, 330k to GND. I've also deleted the decoupling ceramics from both rails which were feeding supply noise into the local GND, so the opamp decoupling now is being handled by the big 'lytics only. Audible upshot of these changes is the sibilance is down to the level I normally expect from my phone with its ES9018. In other words, more lispiness than real sibilance. I will need a cleaner source to verify if its gone totally - a TDA1387-based DAC.
Stage 3 improvement - input buffer. I'm using a current source fed PNP darlington arrangement to buffer the input prior to the volume control. The parts cost of this is negligible as its just 8 two-a-penny discretes and Rs and Cs. Pic attached - the small green postage stamp is the twin buffers.
And how about the sound after this series of tweaks? In a word, stunning. Its jolly hard to stop listening to this for its effortless dynamics. I shall buy a few more to apply these 'enhancements'.
Stage 4 - I was curious as to how good this amp just used as a buffer could make my smartphone sound playing through speakers. The soundstage depth wasn't that impressive (something not noticeable on headphones) so I've now tweaked the power supply a little with a series ferrite bead with 3 turns right out of the boost converter and some more ceramics on the rails. I've also given the opamp its own RC filtering network (2.2ohms and 2 * 10uF 0805 X5Rs). This has given some depth, I'm not sure if any improvement is going to be noticeable on cans though....
Stage 5 - my student has the amp with the first generation set of mods, I gave her two amps to choose from, one modded and the other stock but indistinguishable from the outside. Just listening from her iPhone, she picked the modded one unflinchingly as clearer in about half a minute. The second one I've spent some more time in decoding the fine details of the design. In the process I have noticed what looks to be a design error in that two transistors (the first stage of NPN of both diamonds, Q6 and Q8) have C and E swapped over. I have fixed this but also was curious to explore the rail splitter so I ran a sim with it in LTspice. The PSRR sucks pretty badly at LF so this model has undergone fairly extensive hacks to delete the rail splitter (which also wasted significant power because of the two 1k resistors across the rails) and install output capacitors.
More details on fixing the transistor pin-out errors. I've added a shot of the diamond buffer output part of the PCB, on which you'll see I've already removed Q6 and Q8, the two transistors with C and E swapped over. I'm in the process of cutting the tracks - if you look closely I've already sliced through the collectors and will put a link across to the adjacent via. Then a wire will have to go around to re-connect the E.
In the adjacent picture I've scraped off some solder resist and tinned the copper on the E connection, with a break of the track to the north of the pin. Also added a 0ohm 0603 link to bridge the C connection to the rail.
I've now finished modding the third one and the mod implementations are a little bit more robust. This time around I didn't reduce the supply voltage (still 11.9V), instead using 220uF/16V caps on the main rail (5 in parallel). The higher operating voltage shifts the bias in the output stage so a couple of resistor values needed tweaking. Pic attached. This amp sounds so much cleaner than my headphones (both Superlux and AKG) that now I need to go in search of some upgraded cans....
First the soundstage air was less apparent. Second there's some sibilance noticeable on voices. And third the background hiss is slightly more apparent and a slight whine comes from the power supply. So I figured - open her up.....
Inside its fairly simple, the more or less standard configuration of a pot, then opamp gain stage then discrete diamond buffer. Which is great because I already have experience with this topology. The power supply is a built in LiIon cell with a boost converter supplying 12V in a single rail and there's a passive rail splitter. The dual opamp is an EL2244, one I've not seen before in such a setup.
First up - the input pot is too low a value at 10k, smartphones really only shine when loaded very lightly (at least mine does) - by which I mean at least 200k. I doubt I'll be able to find a replacement pot above 100k so the first requirement here is to give the unit an input buffer. This could be discrete or another opamp - I might try out a low power JFET opamp to begin with. But I'll need to improve the power supply first so let's examine that.... stay tuned.
If you'd like to buy one and don't have access to Taobao, its available ($50, free shipping) from Aliexpress - https://www.aliexpress.com/item/High-...208.3.2.25Nzu6
Update - I did a quick little experiment to see what the effect on the sound of the 10k loading is. To isolate this I used my 'Buffalito' as a buffer (its input impedance is above 200k) and then routed its output into the XuanZu. The result was that the soundstage bloom came back, but the other two problems (sibilance and extra noise) were still there. So it looks like to get the best ambience retrieval the high input impedance is crucial.
Stage 1 improvement - remove the SMT 'lytic caps (100uF/16V) and reduce the power supply voltage to allow 10V caps to be used. I did this by paralleling a 1.1M resistor over the top of the 270k which sets the output of the 5pin boost regulator. Previously the supply was about 12V, now it measures 9.83V so I can use 10V caps with impunity. Having removed similar caps on other portable devices I was expecting a poor ESR measurement and I wasn't disappointed. About 0.8ohm at 10kHz which I figure is the reason for the supply whine.
Of the four caps originally present, two are supply line splitters and the other two are paralleled as reservoirs for what was a 12V supply. I've now fitted 470uF/10V caps but these have ESR more than 10X lower than those I removed. A quick listen shows the supply line whine is gone - half of one problem fixed. Next up, checking the opamp....
Stage 2 improvement - replacing the opamp. The EL2244 is bipolar and runs about 5mA per side. Since the device's output stage is classAB and I want to minimize the noise generation effects of this I changed the opamp to a JFET type so the feedback resistors can be substantially raised in value. Originally there was 10k feedback, 5k6 to GND. Now I have 470k feedback, 330k to GND. I've also deleted the decoupling ceramics from both rails which were feeding supply noise into the local GND, so the opamp decoupling now is being handled by the big 'lytics only. Audible upshot of these changes is the sibilance is down to the level I normally expect from my phone with its ES9018. In other words, more lispiness than real sibilance. I will need a cleaner source to verify if its gone totally - a TDA1387-based DAC.
Stage 3 improvement - input buffer. I'm using a current source fed PNP darlington arrangement to buffer the input prior to the volume control. The parts cost of this is negligible as its just 8 two-a-penny discretes and Rs and Cs. Pic attached - the small green postage stamp is the twin buffers.
And how about the sound after this series of tweaks? In a word, stunning. Its jolly hard to stop listening to this for its effortless dynamics. I shall buy a few more to apply these 'enhancements'.
Stage 4 - I was curious as to how good this amp just used as a buffer could make my smartphone sound playing through speakers. The soundstage depth wasn't that impressive (something not noticeable on headphones) so I've now tweaked the power supply a little with a series ferrite bead with 3 turns right out of the boost converter and some more ceramics on the rails. I've also given the opamp its own RC filtering network (2.2ohms and 2 * 10uF 0805 X5Rs). This has given some depth, I'm not sure if any improvement is going to be noticeable on cans though....
Stage 5 - my student has the amp with the first generation set of mods, I gave her two amps to choose from, one modded and the other stock but indistinguishable from the outside. Just listening from her iPhone, she picked the modded one unflinchingly as clearer in about half a minute. The second one I've spent some more time in decoding the fine details of the design. In the process I have noticed what looks to be a design error in that two transistors (the first stage of NPN of both diamonds, Q6 and Q8) have C and E swapped over. I have fixed this but also was curious to explore the rail splitter so I ran a sim with it in LTspice. The PSRR sucks pretty badly at LF so this model has undergone fairly extensive hacks to delete the rail splitter (which also wasted significant power because of the two 1k resistors across the rails) and install output capacitors.
More details on fixing the transistor pin-out errors. I've added a shot of the diamond buffer output part of the PCB, on which you'll see I've already removed Q6 and Q8, the two transistors with C and E swapped over. I'm in the process of cutting the tracks - if you look closely I've already sliced through the collectors and will put a link across to the adjacent via. Then a wire will have to go around to re-connect the E.
In the adjacent picture I've scraped off some solder resist and tinned the copper on the E connection, with a break of the track to the north of the pin. Also added a 0ohm 0603 link to bridge the C connection to the rail.
I've now finished modding the third one and the mod implementations are a little bit more robust. This time around I didn't reduce the supply voltage (still 11.9V), instead using 220uF/16V caps on the main rail (5 in parallel). The higher operating voltage shifts the bias in the output stage so a couple of resistor values needed tweaking. Pic attached. This amp sounds so much cleaner than my headphones (both Superlux and AKG) that now I need to go in search of some upgraded cans....
Total Comments 0
