I used 50v for the input caps. I think they were .30 cents each at Parts Connexion.
I first replaced the input caps at LINP and RINP (see KJA's photo in post 103), but later also replaced the caps at LINN and RINN.
I am not sure if replacing the LINN/RINN caps made much difference, if any. At this price I might just replace them anyway.
Keep in mind that I am utterly winging everything and do not possess the deep knowledge of others on the board. Just another internet bozo with a soldering iron....
I first replaced the input caps at LINP and RINP (see KJA's photo in post 103), but later also replaced the caps at LINN and RINN.
I am not sure if replacing the LINN/RINN caps made much difference, if any. At this price I might just replace them anyway.
Keep in mind that I am utterly winging everything and do not possess the deep knowledge of others on the board. Just another internet bozo with a soldering iron....
The 3116 has a balanced input that works with the differential output (a good way to get more voltage swing out of a single supply) and a balanced architecture so it can cancel common-mode EMI, RFI etc...
The +/- caps should be the same so the inputs 'see' the same thing and stay in good ... balance.
Many amps also have balanced inputs because many DACs have a differential signal (easy way to get 2Vrms output out of a 5Vdc supply)
The input works best when both inputs are very similar. Looking back from the + input and through the volume pot, when the pot is at full or zero it looks like gnd. That's balanced with the - input because it's connected to gnd.
But when the pot's in-between and worse case is at 1/2 resistance (~3 o-clock) when the positive inputs sees 1/2 of the pot going to gnd and the other 1/2 going to Vin (ac gnd). So for a 20k pot, it will see two 10k resistances in parallel or 5k. This gets added to the input impedance but only the + input.
At 26db gain, the input impedance of the 3116 is 30k. But now the - input sees 30k and the + input sees (30k + 5k =) 35k Not balanced.
So the best way to drive this amp would be from a preamp--which has low output impedance that doesn't add much to impedance seen by the + input ... and with the pot removed.
Last edited:
A good candidate would be a pair of TI DRV135s to drive the amp inputs thereby isolating the amplifier chip completely from the input. If placed very near the input on the board all induced CM noise that invariably will occur with a quite powerful output running through large inductor would be canceled.
I'd stay away from cheapo tone controls. Usually a worse circuit for the amp, and don't solve any problems. Even a 7-band EQ is pretty limited, adjustments on a 2-band one will not please ears.
Well, it doesn't satisfy mine, at least. The "bass" usually affects high bass/low midrange(150-300Hz), making the sound completely bloated, and the highs a low-Q between 7-10kHz, making everything pretty irritating. that's it for the headphone amplifier I have in hand.
Back on the TPA, the YJ boards I've ordered over 2 months ago did not arrive. Most likely lost along the way. Seller asked 10 more days before a refund. Luckly aliexpress seems to cover (or enforce) that if needed.
And so 4 of the taobao boards will soon be on the way. Using an agent to ship here, the fee is small (USD 3) and total price (double shipping is needed) will be under 50 dollars.
Well, it doesn't satisfy mine, at least. The "bass" usually affects high bass/low midrange(150-300Hz), making the sound completely bloated, and the highs a low-Q between 7-10kHz, making everything pretty irritating. that's it for the headphone amplifier I have in hand.
Back on the TPA, the YJ boards I've ordered over 2 months ago did not arrive. Most likely lost along the way. Seller asked 10 more days before a refund. Luckly aliexpress seems to cover (or enforce) that if needed.
And so 4 of the taobao boards will soon be on the way. Using an agent to ship here, the fee is small (USD 3) and total price (double shipping is needed) will be under 50 dollars.
Thanks for sharing Hammett, very interesting.
I've made some observations about PSU. I gathered several types of them, namely 12V 1.5A switching plug type, 12V 3.25A switching notebook type, 12V 1A regulated (7812), 22,5V 3.25A switching from iRobot and finally bench regulated 0-30V 5A. Listening test (I used Youn Sun Nan tracks #1 "Dancing With You" and #8 "Shenandoah" from Voyage album - they are well recorded, great for details and emotions, #1 has some nice clean bass, and both are enjoyable to listen to endlessly )) shows the following:
1.All 12V units perform very closely, giving pleasant smooth transparent sound.
2.22,5V PSU gives less transparent, non-involving, fatiguing sound, "stressed" might be proper definition.
3. Bench supply allows for some expirementation. It confirmed previously mentioned in the thread that voltage above 21V making sound worse (so this is not related to PSU type). Even more, lowering voltage further there is slight increase in transparency, female voice details and emotions getting better defined. While 20V is pretty fine, best of all it seems for me sounding on voltages below 10V (pretty well performs even on 6V).
Also own amp noise is noticably lower on lower voltages, I'd say about 6 dB difference between 24V and 6V.
Regarding power requirements, bench supply meter showed board consuming very small current. On my typical listening levels (and you might get the idea of music - mostly vocal jazz and ECM) current draw is as low as 30 mA! and that @ 6V (one might think it could me even less when increasing voltage for same power but no, at 24V it draws about 40mA). So in my case basically all current is consumed by the chip circuitry itself not speakers as they are sensitive enough (Alpair 7.3). Suprisingly low though.
In search of possible high voltage problem I've made measurements of voltage offset after input caps to get an idea about input amp circuitry. It showed an interesting thing - as power voltage rise, DC offset decreases (of coarse it is same on all inputs). Here's the table (offset/power voltage):
2.8 @ 6
2.6 @ 9
2.3 @ 12
1.8 @ 18
1.3 @ 24
This observation may led to think that sound could get stressed even in input stage due to nonlinearities at lower voltages. There is also some room left for experiments with gain setting though as it could be PGA design fault. Surely it may depend on inductors also, but' I'm not up to it yet.
So I decided to stay with 12V linear PSU as it seemed tad better of them all and have enough power for my even 'loud' listening.
I've made some observations about PSU. I gathered several types of them, namely 12V 1.5A switching plug type, 12V 3.25A switching notebook type, 12V 1A regulated (7812), 22,5V 3.25A switching from iRobot and finally bench regulated 0-30V 5A. Listening test (I used Youn Sun Nan tracks #1 "Dancing With You" and #8 "Shenandoah" from Voyage album - they are well recorded, great for details and emotions, #1 has some nice clean bass, and both are enjoyable to listen to endlessly )) shows the following:
1.All 12V units perform very closely, giving pleasant smooth transparent sound.
2.22,5V PSU gives less transparent, non-involving, fatiguing sound, "stressed" might be proper definition.
3. Bench supply allows for some expirementation. It confirmed previously mentioned in the thread that voltage above 21V making sound worse (so this is not related to PSU type). Even more, lowering voltage further there is slight increase in transparency, female voice details and emotions getting better defined. While 20V is pretty fine, best of all it seems for me sounding on voltages below 10V (pretty well performs even on 6V).
Also own amp noise is noticably lower on lower voltages, I'd say about 6 dB difference between 24V and 6V.
Regarding power requirements, bench supply meter showed board consuming very small current. On my typical listening levels (and you might get the idea of music - mostly vocal jazz and ECM) current draw is as low as 30 mA! and that @ 6V (one might think it could me even less when increasing voltage for same power but no, at 24V it draws about 40mA). So in my case basically all current is consumed by the chip circuitry itself not speakers as they are sensitive enough (Alpair 7.3). Suprisingly low though.
In search of possible high voltage problem I've made measurements of voltage offset after input caps to get an idea about input amp circuitry. It showed an interesting thing - as power voltage rise, DC offset decreases (of coarse it is same on all inputs). Here's the table (offset/power voltage):
2.8 @ 6
2.6 @ 9
2.3 @ 12
1.8 @ 18
1.3 @ 24
This observation may led to think that sound could get stressed even in input stage due to nonlinearities at lower voltages. There is also some room left for experiments with gain setting though as it could be PGA design fault. Surely it may depend on inductors also, but' I'm not up to it yet.
So I decided to stay with 12V linear PSU as it seemed tad better of them all and have enough power for my even 'loud' listening.
Last edited:
The treble control is just a crude top cut that shunts a 0.1µ capacitor in series with the pot and a 1k resistor to ground at the input of the left and right power amplifiers. It does not provide any boost at all and the amount of cut depends on the setting of the volume control.
I have disabled mine by removing the 1k resistors.
The L.P.F. for the the sub channel is 72Hz, no H.P.F is provided to the L and R channels. I reduced value of the coupling capacitors to the left and right amplifier channels to remove some of the low frequency content until I get round to adding a proper filter. There is no point in having a flat response to below 10Hz unless you need to rattle the windows.
Excellent review of power supply effects! Thank you.
So you actually think it sounds better at 12v vs 19v?
I have a few comments on this
I assume that's at the same relative listening level which actually means the chip has higher own noise at lower voltages relative to the maximum output as that is 12dB higher at 24V than at 6V.
The power consumption is completely dependent on load, modulation type, filter type and so on, so it's difficult to say anything definitive. However, when muted, ie. the outputs are not switching, the chip power consumption scales almost perfectly with input voltage from 8mA@6V, over 15mA@12V, to 28mA@24V. These are rough figures and are almost independent of modulation type and any other settings on the chip.
The increased losses are most likely incurred by the internal voltage regulator that supplies all the input and control circuits on the chip. By reverse engineering those figures you can assume the total power consumption of the chip that are not related to output switching losses or the voltage regulator is about 3mA regardless of supply voltage.
This to me confirms the previous observation that own noise of the chip is higher at lower voltages relative the maximum output. It also to me, at least, indicate that the culprit of the sound quality issues at the highest voltages, is, as I previously stated, to be found in the output transistors on the chip.
EDIT: I will also note that the TPA3118 when properly heat sinked to a PCB with high copper thickness is tolerant of slightly higher voltages before sound quality issues occur than the TPA3116 when that is also properly heat sinked to an aluminium heat sink. This is most likely due to the higher thermal transfer rate of copper and that the chip substrate is closer to the thermally conducting surface on the TPA3118 than on the TPA3116.
I assume that's at the same relative listening level which actually means the chip has higher own noise at lower voltages relative to the maximum output as that is 12dB higher at 24V than at 6V.
The power consumption is completely dependent on load, modulation type, filter type and so on, so it's difficult to say anything definitive. However, when muted, ie. the outputs are not switching, the chip power consumption scales almost perfectly with input voltage from 8mA@6V, over 15mA@12V, to 28mA@24V. These are rough figures and are almost independent of modulation type and any other settings on the chip.
The increased losses are most likely incurred by the internal voltage regulator that supplies all the input and control circuits on the chip. By reverse engineering those figures you can assume the total power consumption of the chip that are not related to output switching losses or the voltage regulator is about 3mA regardless of supply voltage.
This to me confirms the previous observation that own noise of the chip is higher at lower voltages relative the maximum output. It also to me, at least, indicate that the culprit of the sound quality issues at the highest voltages, is, as I previously stated, to be found in the output transistors on the chip.
EDIT: I will also note that the TPA3118 when properly heat sinked to a PCB with high copper thickness is tolerant of slightly higher voltages before sound quality issues occur than the TPA3116 when that is also properly heat sinked to an aluminium heat sink. This is most likely due to the higher thermal transfer rate of copper and that the chip substrate is closer to the thermally conducting surface on the TPA3118 than on the TPA3116.
Last edited:
Input was muted at DAC, vol pot is at the same position so we are talking solely to noise vs power voltage. While max power output is about clipping not about gain, gain remains the same being powered from 24V or 6V. So I'm afraid there is no practical sense in this and therefore noise is still remains less with lower power voltage, not higher
Relation between noise differencies, DC offset change and transistors is still unclear for me, would you please explain your idea further?
I'd also addreess voltage tolerance differences between 3116 and 3118 not to termal effects as sound change is observed while chip is absolutely cold but merely to sample deviation (which is more likely and probably support my PGA becoming nonlinear theory). People here talking about different thresholds for 3116 btw and one with several could check this. I've got only two of them and not that interested as getting these amount of power makes no sesnse for me
Last edited:
Input was muted at DAC, vol pot is at the same position so we are talking solely to noise vs power voltage. While max power output is about clipping not about gain, gain remains the same being powered from 24V or 6V. So I'm afraid there is no practical sense in this and therefore noise is still remains less with lower power voltage, not higher
You missing the point. The noise is lower at lower voltages but lower at higher voltages relative to the maximum output. That means the noise does not scale proportionally with the supply voltage. If own noise was the same at all supply voltages it would have to be 12dB higher at 24V than 6V, not 6dB.
It's relatively simple as it's precisely the same effect as increasing bias current in a normal class AB amp; ie. higher bias, lower noise, and lower dc offset (again relative to maximum as above).
Absolutely cold heat sink is by no means a precise measurement of internal substrate temperature variations as we're talking about differences that would be less than a few 1/100th of a degree on the heat sink
Last edited: