So far very pleased with the sonic results of this amp, it seems to have a slightly lower noisefloor than my TDA8932-based monoamp which manifests as improved saturation of tonal colours.
I needed a way of testing the TDA1521s as mine are all recycled ones, so now there's this single channel board -
It still retains the capability of being paralleled. It can be used alone to create a mono ~20W amp running on +20V so long as the speaker impedance isn't lower than 8R. Two paralleled will give twice the output current and allow a higher supply voltage, allowing in the region of 60W into 8R.
I needed a way of testing the TDA1521s as mine are all recycled ones, so now there's this single channel board -
It still retains the capability of being paralleled. It can be used alone to create a mono ~20W amp running on +20V so long as the speaker impedance isn't lower than 8R. Two paralleled will give twice the output current and allow a higher supply voltage, allowing in the region of 60W into 8R.
@abraxalito Richard, maybe still early days, but what is coming in the future for this?
Will you make boards/kits available for members here? If yes i will be watching this.
Will you make boards/kits available for members here? If yes i will be watching this.
Having gotten good results with the buck regulator module, I ordered up a couple more for building a second prototype amp. Turned out one of those didn't work properly, the output voltage couldn't be set stable, it gradually sank over tens of seconds and then suddenly jumped up with a click emanting from somewhere within. So I thought about sending it back. In the meantime though I connected up the second one to the toroid and its associated rectifier/smoother board which wifey kindly designed.
The off-load DC voltage coming out of the rectifier board was 69.2V which is comfortably below the max 75V for the buck module so I hooked them together and all was dandy. I did notice though a bit later that the toroidal inductor was hot to the touch, even under no-load conditions. So perhaps the optional fan isn't just for high power use, high voltage may need it too. Seeing as I had one dead module on my hands I decided to take it apart to see what the inductor value was and whether I could create a lower loss variant. In the process of disassembly it became apparent the reason this module didn't work - one of the MOSFET pins (the gate) was dangling in mid-air and had never made contact with the PCB because it was bent forward. I figured I could fix that. I removed the inductor and checked its value - 43uH. It was wound with two strands of 0.8mm dia wire - on the LCR meter the AC losses looked significant (0.2R or so) even though the DCR was in the 10mohm region. I decided to try rewinding it with some homebrewed Litz wire. The inductance though came out lowish, around 35uH, but I went ahead and plugged it in.
The new inductor works great in the previously defunct module and now is only slightly warm to the touch. Flushed with the feeling of double success, I decided to make another one with slightly higher inductance (closer to the original) wound onto a 0.9" sendust core I had in my box of ring cores. This didn't turn out so well as it also ran rather too hot. I wonder should I suspect my sendust core isn't really sendust or perhaps the buck module was using something a step above sendust? I checked the loss of my core with the LCR and sure enough it was significantly higher than that of the core supplied with the buck. I'll check out whether there's any core with lower loss available in due course - perhaps 'hi-flux' will fit the bill?
Original winding on the left, home-made litz wire on the right.
The off-load DC voltage coming out of the rectifier board was 69.2V which is comfortably below the max 75V for the buck module so I hooked them together and all was dandy. I did notice though a bit later that the toroidal inductor was hot to the touch, even under no-load conditions. So perhaps the optional fan isn't just for high power use, high voltage may need it too. Seeing as I had one dead module on my hands I decided to take it apart to see what the inductor value was and whether I could create a lower loss variant. In the process of disassembly it became apparent the reason this module didn't work - one of the MOSFET pins (the gate) was dangling in mid-air and had never made contact with the PCB because it was bent forward. I figured I could fix that. I removed the inductor and checked its value - 43uH. It was wound with two strands of 0.8mm dia wire - on the LCR meter the AC losses looked significant (0.2R or so) even though the DCR was in the 10mohm region. I decided to try rewinding it with some homebrewed Litz wire. The inductance though came out lowish, around 35uH, but I went ahead and plugged it in.
The new inductor works great in the previously defunct module and now is only slightly warm to the touch. Flushed with the feeling of double success, I decided to make another one with slightly higher inductance (closer to the original) wound onto a 0.9" sendust core I had in my box of ring cores. This didn't turn out so well as it also ran rather too hot. I wonder should I suspect my sendust core isn't really sendust or perhaps the buck module was using something a step above sendust? I checked the loss of my core with the LCR and sure enough it was significantly higher than that of the core supplied with the buck. I'll check out whether there's any core with lower loss available in due course - perhaps 'hi-flux' will fit the bill?
Original winding on the left, home-made litz wire on the right.
Hi. Maybe a little offtopic, but i would like to share my experience with such inductors , used in psu and converters. So , recently purchased a smps board , rated 24V 12A ,15A max. Same problem , inductor was hot while psu was loaded. And heating problem was not inductor wire resistance , hot was a ferrite itself. It was kinda of core ,used in atx supplies ,but black color. I made experiment and removed it , in place of it placed a EI core with some air gap, used few layers of insulation tape to make gap. That was a flyback type psu from old tv. I wound winding with single 2,5mm2 flexible wire from cable. And you know what ? Heat in inductor dissapears! Also , have performed efficiency tests , output voltage and load current was monitored, also consumption from mains and recorded. A few wats were saved easily , and this makes my think , that inductor in step down converters ,or current doubler in forward converters, do require air space to not saturate , also they like transformer core need to be sized properly... Most china used inductors are too small , and while this saves space and cost, efficiency is not greatest possible. One of the best ring type inductors , working properly in step down converters, have dark blue color ferrite ring ,and do not saturate so easily. DTMSS-40 https://www.tme.eu/en/details/dtmss-40_0.047_45v/toroidal-inductors/feryster/dtmss-40-0-047-45-v/ as example, may be useful. I was unable to locate this core separately.
Likely that core was Sendust. Thanks for your comments, gapped ferrite is going to be lower loss than Sendust, as far as I'm aware ferrite is the lowest loss inductor.
I've had one of the buck modules fail after working fine for a few weeks, so I took it apart:
The SOIC-8 at the bottom I have identified as XL7005A, it's supplying the rail for the adjacent SOIC-16 which is still a mystery component. What failed is the schottky diode (D2) so gotta order some S210 from Taobao.
So-16 component can be TL494 or it's clone KA7500, if pinout matches. Failed diode probably heated and had no heatsink, maybe smd ? There exists similar module, if i remember right, 8 Ampere rated, with LT circuit, which heats very low and have current limit ability. But it's input voltage is up to 60V , so it may not fit your design.
Good call, TL494 does seem to match up, thanks. Not sure how that diode would be overheating, could be they fitted one with too low a volt rating. It got destroyed on removing it so I can't investigate.
60V is too low a max rating in this application with 69VDC coming out of the rectifier board.
60V is too low a max rating in this application with 69VDC coming out of the rectifier board.
On putting the buck module inside the case with the amps I got a noise problem which seemed to be originating from said buck module. After poring a lot over the circuit of the disassembled unit, I discovered that the heatsinks are left 'flapping in the breeze' electrically speaking and there's a large fill area under the low-side MOSFET heatsink which is connected to its drain terminal (that's the large rectangular white area on the right in the pic above). Perhaps the designer was concerned about capacitance between FET and heatsink? Seeing as that FET's drain the noisiest voltage node in the whole show and its being given several cm2 of radiating opportunity I decided major surgery would be needed - to have that heatsink tied firmly down. With the Stanley knife I hacked away at the edges of the fill and liberated it from the source of noise.
Looking at the output residual there was the familiar triangle wave which is characteristic of a buck converter but there were also some other weird contortions of the slopes which, it eventually became clear were due to interference from the XL7005A buck converter. The grounding of the schottky in a non-sync buck is hyper-critical for noise on the ground, can't find the application note right now where I learned this. On this layout the grounding of that diode isn't to the XL7005 GND pin (and hence to the switch) but goes all around the houses before returning to the controller chip. Fixing that up got rid of the weird notches that were showing up superimposed on the triangle waveform.
The other issue I noticed was that in stock form the ripple rejection was looking very poor, only something like 5X. I guess if you feed this buck from an already regulated supply (or battery) you'll never notice this, however here the input is coming from rectified 48VAC so a modicum of ripple rejection is rather important. Turned out one of the error amps in the TL494 was being slugged way too hard (100nF feedback cap) so even at 100Hz there was hardly any loop gain available. I reduced this cap to 1nF but that led to instability when fed from high input voltage (though not from my bench supply) with squealing noises from the inductor and rapid overheating. In the end 3n3 worked OK, in conjunction with the XL7005 grounding mods.
Looking at the output residual there was the familiar triangle wave which is characteristic of a buck converter but there were also some other weird contortions of the slopes which, it eventually became clear were due to interference from the XL7005A buck converter. The grounding of the schottky in a non-sync buck is hyper-critical for noise on the ground, can't find the application note right now where I learned this. On this layout the grounding of that diode isn't to the XL7005 GND pin (and hence to the switch) but goes all around the houses before returning to the controller chip. Fixing that up got rid of the weird notches that were showing up superimposed on the triangle waveform.
The other issue I noticed was that in stock form the ripple rejection was looking very poor, only something like 5X. I guess if you feed this buck from an already regulated supply (or battery) you'll never notice this, however here the input is coming from rectified 48VAC so a modicum of ripple rejection is rather important. Turned out one of the error amps in the TL494 was being slugged way too hard (100nF feedback cap) so even at 100Hz there was hardly any loop gain available. I reduced this cap to 1nF but that led to instability when fed from high input voltage (though not from my bench supply) with squealing noises from the inductor and rapid overheating. In the end 3n3 worked OK, in conjunction with the XL7005 grounding mods.
Last edited:
So then probably phase mismatch occured, or frequency dependent amplitude mismatch, and one chip taken all the load. To prevent this, another chips should be used, when only output stages paralleling, and current summing , like tda7293/94. Or maybe layout dependent oscillation occureda, when prototype was fine. Of course if chips were from reputable source
It seems clear that one chip took a much bigger proportion of the load than designed for. My hypothesis for the way this happened though isn't any of your suggestions. What probably happened is a kind of thermal runaway due to the fact that I'm running the chips in lower gain than they're designed for.
When run in standard configuration (30dB) both the feedback and input resistors are on-chip. With lower gain (23dB), I add an extra resistor in series with the input resistor and trim it to get good matching. The trouble is - this external resistor doesn't have a significant temperature coefficient whereas the two on-chip resistors do. In stock form, the tempcos of the on-chip resistors don't matter because the ratio of those two resistors determines the gain and they track together. However seeing as roughly half the total of the input resistance is off-chip its effective tempco becomes only half that of the feedback resistor with the result that the two tempcos no longer cancel. The gain therefore varies significantly with temperature and for paralleled amps this is a disaster waiting to happen.
When run in standard configuration (30dB) both the feedback and input resistors are on-chip. With lower gain (23dB), I add an extra resistor in series with the input resistor and trim it to get good matching. The trouble is - this external resistor doesn't have a significant temperature coefficient whereas the two on-chip resistors do. In stock form, the tempcos of the on-chip resistors don't matter because the ratio of those two resistors determines the gain and they track together. However seeing as roughly half the total of the input resistance is off-chip its effective tempco becomes only half that of the feedback resistor with the result that the two tempcos no longer cancel. The gain therefore varies significantly with temperature and for paralleled amps this is a disaster waiting to happen.
I've modified the gain for all the chips to 30dB now and because that's a little too high, I stole a couple of dBs back again by resistively loading the trafo (rather than use an RC load). So now overall the gain is about 26dB which is only 3dB higher than before.
Its going into this $23 aluminium case, some pics will come along later : https://www.aliexpress.com/item/1005004717423123.html
Its going into this $23 aluminium case, some pics will come along later : https://www.aliexpress.com/item/1005004717423123.html
I only saw this thread recently, and was thinking about how to reduce gain in the audio band while still having higher gain for stability. Suppose we use a resistor value R between the output and negative input to lower gain to the desired amount. Split it into two series resistors, each R/2, and have a capacitor between their connection and ground. The capacitor value would be set with r/2 to roll off above 20kHz, restoring full gain at higher frequencies and giving the desired stability.
Achieving matched values for three sets of this circuit is left as an exercise.