Improving TPA3118 Class D Amplifier

jls001

Well-Known Member
Joined
Feb 9, 2010
Messages
6,180
Points
113
Location
Mumbai
Intrigued by the high interest that the compact amplifier boards based on the family of Texas Instruments' TPA 3118/3116/3110 of Class D amplifier chips have been generating among the enthusiasts community worldwide, I decided to find out for myself what these boards sound like stock, and if the many modifications others have tried really do improve performance to a higher level.

This is the board I got, though mine was from eBay India. Each is a mono board so one needs two for stereo.

Below is an actual photo of one of my boards. They're very compact and measure roughly the same size as a credit card.

1ZTuRYU.jpg


There are many vendors selling similar boards on Amazon, eBay, Alibaba, Taobao, etc and there are quite a few variants to choose from. Just to give an idea of the variety, you will find stereo boards, stereo boards with volume pot, stereo boards with Bluetooth, and plain PBTL mono boards with double the power output of the stereo boards. The TPA3116 boards usually have some form of heat sink placed on top of the chip itself. Due to its higher heat dissipation ability, it can make more power compared to TPA3118 based boards. TPA3118 has heat transfer pad at the bottom of the chip and is supposed to transfer the heat to the circuit board itself, with the circuit board itself acting as the heat sink. Typical stereo TPA3118 boards produce about 30 Watts into 8 Ohm load with 24V supply.

The stereo boards typically come with a gain of 36 dB, making it quite suitable for plugging in portable players or smart phones directly. My boards came with a slightly lower gain of 32 dB. My understanding is this gain of 32 dB is still higher than the more typical 26-28 dB gain on most power amplifiers. The downside of higher gain is the higher noise floor.

If one wants to understand more about this chip amp, it is essential to go through the datasheet here. It gives all information needed by board designers, including example schematics on which most of the commercially available boards are based on. There is also information on how to change the gain of the amplifier.

The stock board is actually a very nice sounding amplifier. I drove it with my Kuartlotron buffer, the buffer acting as the volume control. The output of the TPA3118 board drives my 8 Ohm speakers. One can almost live with the sound of the stock board, except for a hardness in the midrange. This leads to a harshness and makes it difficult to listen for extended time. Though the datasheet example circuit specifies a load of 4 to 8 Ohms, it is my understanding from reading many hundred pages of forum posts that the output filter inductor of 10 uH is actually more suited to 4 Ohm load, and can be stretched to 6 Ohms. So if you want to drive an 8 Ohm load like me, the most fundamental change one must make is to swap out the inductors to the more correct value. I have variously come across 15 uH, 22 uH and 33 uH as being the best value for 8 Ohm speakers. I have not undertaken this mod as I am not sure about the value, and also because the cost of replacement inductors is going to be more than the cost of the board itself.

There are cheaper mods that will bring substantial improvements. The first and foremost is to change the four 330 uF/25V power decoupling capacitors to something better. I used 470 uF/35V Keltron caps. The replacement caps must be 10 mm diameter or less to fit within the available space, and rated for at least 25V. Changing to better capacitors completely removed the harshness that I heard. The stock caps are surface mount type. Care must be exercised when desoldering so as not to lift the pads. Caps with regular leads can be used as replacement by bending the leads carefully to match the pads on the board. This mod is the most important mod, in my opinion.

Next on the agenda was to decrease the gain from 32 dB to about 26 dB, mainly because I can hear a whining noise from one channel, and also because I was not happy with such a high gain. There are two resistors whose values can be changed as per the gain table given in the datasheet for setting the gain to 36, 32, 26 and 20 dB. The easiest to set is 20 dB because it involves removing one of these resistors from the circuit board. I wanted to set mine to 26 dB as that is closest to what I understood to be the gain of a conventional power amplifier (of course there are low gain power amplifiers like Firstwatt F5). This involved removing one of the two resistors mentioned above and replacing it with appropriate value. I am avoiding going into details like which R, which value, etc as it gets fairly technical, but if someone is interested, I can provide more info on how to go about it. But be warned that desoldering SMD resistors is not easy. Soldering back the replacement resistor is tougher, especially for someone like me with no prior experience of working with SMD parts. It's perfectly possible to brick your board. I very nearly rendered one board useless while desoldering the resistor as the PCB trace got lifted and fell off. Choosing your gain depends on how much gain you have at the source-preamplifier stages. If you have a 10-12 dB preamp stage, even 20 dB at the power amplifier stage (which this TPA3118 board is) ought to be sufficient. I would not use the highest gain setting of 36 dB as the noise also gets amplified by that amount. The next lower gain of 32 dB is also high.

The third mod one can do (with some caveat) is to beef up the power supply rail. I use 19.5V/4.7A laptop (one per board). I used a 6800 uF/50V capacitor across the 19.5V DC supply into the board (+ve terminal of capacitor on +ve terminal of the supply, -ve terminal of capacitor on -ve terminal of the supply). I use two laptop bricks from two different brands, though they have exactly the same voltage and current ratings. One channel powered by brand X adapter had been consistently having a high frequency whine audible even when the volume control is zero. The volume of the whining noise does not increase even when volume knob is turned up. I tried a simple R as well RC filters across the source RCA socket but neither had any effect on the whining noise. A couple of days back I noticed that one channel is sounding lower than the other side. I rechecked the values of the resistors I used on both channels and discovered that they were not closely matched. Later, I used closely matched resistors but still had channel imbalance. I reflowed at solder points on the board with the lower gain, in the hope that it could cure a poor solder but that didn't help either. Finally, I took off the power rail capacitor (6800 uF that I had fitted) on both boards, and that instantly cured both the channel imbalance and the whining noise. So in effect I had introduced two problems when trying to beef up the power supply rail. This is very likely because of a degraded capacitor as the other channel worked perfectly. Hence the caveat at the beginning of this para. I may re-use proper CRC at the supply rail when I can buy fresh capacitors. For now I am running without them.

The last mod that I did was to use bootstrap snubbers for the output inductor filters to reduce EMI that can arise by the switching of the output stage (sudden transients into inductive loads produce EMI). The "output" side of the inductors already have snubbers by way of the zobel network. What is added is a C and R before the signal reaches the inductor. This additional snubber cleans up the sound further.

There are further mods like playing around with different switching frequencies, but I don't plan to play around with these boards any more. I am afraid I will brick them as the thickness of the copper pads is thin and cannot withstand multiple soldering/desoldering.

In a coming post, I will post about my impression of the sound after all these mods.
 
Also note that changing the gain impacts the input impedance and by virtue of association, the input high pass filter [formed by the input impedance and the input filter capacitor] so if you plan on reducing gain it would be a good idea to also change the input capacitor to the value given in the table in section: "7.3.2 Input Impedance" of the datasheet.
 
so adding an extra reservoir capacitor increased the whine? That's confusing

was this in addition to the existing capacitor bank or as a replacement?

perhaps try a film cap as a bypass to the big electrolytic?
 
so adding an extra reservoir capacitor increased the whine? That's confusing

was this in addition to the existing capacitor bank or as a replacement?

perhaps try a film cap as a bypass to the big electrolytic?

I think that specific capacitor might be faulty. The other side using the same value is fine. I plan to get some fresh pieces and try once more before I box it up.

This cap I'm referring to is placed off-board and is extra, and not a replacement of the on-board four 330 uF caps (though those have also been replaced with 470uF, then later with 1000 uF).

Yes, one could try a 100 nF or so bypass to the off-board caps.
 
Build pics.

Featureless bottom side of boards:

RhiWQXB.jpg


Note the tiny via holes directly below the chip. That's the thermal coupling from chip to PCB.


Wired up for testing:

bHViIH5.jpg



Boxed up:

LfagpFW.jpg


The two switches on the front panel are Mute switches. There is no LED extendable from the board to the front panel, although the board does have one red SMD LED.


Rear panel:

HDLgpnL.jpg


The switch at the center of the panel is on-off switch for the two separate DC supplies from laptop power adapters. The switch is two pole so both live wires of the two separate DC supplies can be switched on one switch instead of using two separate switches.


Close up of one board:

IX6quqO.jpg



Cost tally:

Each is a mono board so we need two boards for stereo.

Two boards cost Rs 1070 shipped from Bangalore.

It's powered by two old laptop adapters, each 19.5V/4.7A. I bought used laptop power bricks from the local electronics market for 350 each. So add 700.

Add two RCA connectors for audio inputs. About Rs 300.

Add four speaker binding posts. About Rs 200 for four. I used el cheapo connectors in keeping with the theme of keeping the cost down. But I won't use these cheap binding posts in future as one pair didn't even have insulation to insulate from chassis. I can confirm that shorting binding posts to chassis does not work. It has to be insulated. BTW, did anyone noticed the mismatched pair of "socks" on the rear panel?;)

Add about Rs 250 for three switches (two SPST switches for Mute and one DPDT switch for DC on-off switch).

Add about Rs 40 for two DC input sockets. They're cheap plastic body type. I could not get anything better but to their credit they've stood up to lots of soldering abuse without the connectors falling off or the plastic body melting when soldering.

Add Rs 280 for a blank steel box on which we need to drill required holes.

So that's about Rs 2840.

I had mentioned earlier about the whining noise due to the power rail capacitor. I tried fresh capacitors and got the whining back so I'm not going to bother with beefing up the rails. I think i have beefed up the on-board power decoupling capacitors sufficiently (four 1000 uF instead of four stock 330 uF caps).
 
Last edited:
Wouldn't rotating the PCB's 108 deg. be better? Shorter wire for I/P's.
Never underestimate the effect of short wires :D
 
Last edited:
Some of the modding I had done:

First and foremost is to replace the four 330 uF/25V capacitors on the board with something better and higher value. This brought the most improvement as it completely removed the harshness in the stock board.

These four caps are the ones marked "330 25V RVT" as seen on the photo below:

1ZTuRYU.jpg


These caps have surface mounted terminal so it is not easy to rock it from side to side while heating it up to remove it. Care must be exercised to avoid peeling off the circuit board trace. Apply only gentle force while heating with soldering iron. To access the terminals of the capacitors you need needle tipped bit on your soldering iron. The space between the capacitor barrel and the board is very less. My regular 25W Soldron tip could not reach the cap terminals. I used a Goot with needle ceramic bit. Replacement capacitor diameter must be 10 mm or less to fit the available space on the board.


Second change one can make is to reduce gain. My boards come with 32 dB of gain. Here's the table from the datasheet showing the values of resistors:


Z3H3Jba.png



We need to change the values of R1 and R2 given in the circuit diagram. For 32 dB gain in Master Mode, the values of R1 and R2 are 39K and 100K Ohms. Note that the datasheet circuit example uses another set of names for R1 and R2, so it is best to identify them from the nearest pins on the amp chip. On the circuit board (at least mine), R1 and R2 corresponds to R28 and R27. See the first photo in this post for reference. If you remove R27 completely, the table shows that the gain will be 20 dB. This is the easiest path. However, 20 dB gain may be too less for some chains. I preferred 26 dB gain so I removed R28 = 39K and used 20K instead. Later I changed it to 27K for slightly higher gain. Since I don't know where to source SMD resistors, I simply used a regular 1/8W through hole, metal film resistors and soldered them carefully across R28.

Here's a pic of the 10K + 10K resistors in series on the top left corner near the orange capacitor. I didn't have 20K resistor so I had to use two. Later I had changed it to a single 27K:

IX6quqO.jpg


The blue "drop" capacitor riding piggyback on the orange capacitor is for bootstrap snubber, explained below.


Third mod I did was to fit bootstrap snubbers. These are basically one 330 pF capacitor and a 10 Ohm resistor in series, the 10R being grounded. Check out the diagrams below, borrowed from here:


boGTRQV.png



and the closeup below:


v9cSRJg.png



C10, 12, 14, 16 are the bootstrap caps.

C11 and R11 constitutes the snubber for the C10. There is no C11-R11 on the stock board, and that's what we need to install. C11 and C17 are the blue caps visible in the pic above.

Repeat for C12, 14 if you have a stereo board, otherwise ignore them if you have PBTL boards.

So basically for PBTL board you install:

C11-R11 for C10, and C17-R16 for C16.

Here's a diagram on how to implement it practically:

LCMEVcD.png


The 10R resistor is grounded on the negative terminal of the 330 uF cap, since the negative terminal of the 330 uF cap is at ground. One end goes to one lead of the 10 uH inductor. These are two convenient points to solder.

Also note that there is a small red line connecting R27 and R28. I use this since the pad of R28 fell off while desoldering.

The last mod is to put a capacitor of sufficiently high value across the DC input terminals - negative terminal of capacitor to negative of DC supply, positive terminal of capacitor to positive of DC supply. I tried 6800 uF/50V. You can safely use anything rated above 35V. This capacitor is supposed to provide some extra energy storage that can help stiffen the power supply line and prevent the power supply from sagging during high dynamic passages. Or at least that's the theory. I have removed this caps since it led to whining noise in one channel. The other is surprisingly dead quiet.

If you plan on the above surgeries, do be careful as the board traces are very thin and break easily.
 
Wouldn't mounting the PCB's 108 deg. be better? Shorter wire for I/P's.
Never underestimate the effect of short wires :D

Yes, you are absolutely right! The audio input wire ought to be kept shortest. Power cables and speaker cables carry much higher signal so the extra length ought to matter less for them. Anyway, I have boxed it up and screwed down the top. Will follow your advice in next project.
 
I reviewed the ckt briefly and found there are enough capacitors in the power-supply and output stage to reduce hf noise. I can suggest only one additional mod which i dont see. Add one 0.022mfd capacitor between power supply ground and amp chassis. Also use ferritte beads in the input signal cables if possible.
 
Thanks for the suggestion, Hari.

I have 330 pF/400V. Will that work in place of 220 pF you suggested?

Alternately, can I parallel two 0.01 uF/100V? Or series yep 0.047 uF/325V?
 
Last edited:
If you are willing to wait for 20+ days for it to reach you, these boards are available on eBay for ~US$6 with free shipping for 2 boards. BARGAIN!!!
Some months ago, I got them for ~US$3.50 for 2 boards. That was before they became famous on DIYA. :D
 
TPA3118 listening impressions

The initial sound of the TPA3118 with everything stock was harsh in the mids and highs. I didn't have the patience to wait for it to burn in since I had pre-planned the mods I wanted to do. After changing the four 330 uF capacitors on the board to 470 uF Keltrons, the improvement was immediate. I attribute this to the Keltron capacitors because the change was immediate. If one could get hold of better capacitors, it is reasonable to believe that there could be further, if incremental, improvements. What the change of capacitors did was to tame the harshness. It was cured.

Later, I swapped out the 470 uF caps to 1000 uF caps (a brand called CapXon which I'm using for the first time). I didn't hear immediate gains in sound, but now that I've put in more hours on the amplifiers, I think it was a worthy investment.

Since I had only recently completed building MyRef Rev C amplifier based on the LM3886 chip, and been listening to both amps side by side, I can't help but contrast and compare the TPA3118 sound to that of the MyRef Rev C. I'll try and break it down:

Bass weight is where MyRef wins over the TPA3118. MyRef has a mid bass bump which makes a standmount that inherently has limited reach in the bottom end, shine. It imparts it added bass weight. The bass texture though is a bit rounded off, and not too defined, but it certainly makes for pleasant listening. In contrast, the TPA3118 has relatively lighter bass weight and is flatter without the mid bass bump, but it seems to have better defined texture. I feel that on speakers with bigger woofers or bigger cabinet or with careful placement of speakers for bass reinforcement, the TPA3118 bass weight can be definitely improved. I haven't moved my speakers an inch as I had arrived at the present position by repeated incremental changes. In absolute terms, if one were to hear the TPA3118 on its own without reference to another amplifier, one may not find it lacking much in the bass. So the differences are not the night/day sort.

Midrange is where the MyRef really excels. It just sounds so good to my ears that I still grin to myself when listening to it. I'm guessing there is some sort of a bump in the presence region to make voices so euphonic. The TPA3118 on the other hand is more even handed. It's simply flat right across the audio spectrum and its presentation is quite neutral. But here, being neutral doesn't necessarily equate to being boring.

The highs are where the TPA3118 simply outshines the MyRef. The latter, as I reported in the MyRef build thread elsewhere, has a rolled off high. The TPA3118 sizzles and sparkles, without being etched. The operative word is without being etched. One can sit down and listen for hours without being fatigued. The treble after the hours I've put it is very smooth. In its defence, I must add that the MyRef treble may not have the reach of the TPA3118 but the sweet midrange more than makes up for it. One doesn't miss the sizzling highs. This makes MyRef much more forgiving of poorer records which is a good thing as most of us have CDs/records that we love despite the less than stellar sonic qualities.

I have extensively heard various genres on the class D TPA3118 and can confirm that it is a tonemeister - a champion, in fact. It is also very resolving of details. Macro dynamics is another strong point. I am not able to milk the full power of the chips because I'm using lower supply voltage than the recommended 24V, but I certainly don't find it lacking in dynamic capabilities or unable to drive my speakers loud despite using 19.5V supply. It has startling dynamics (when present in the program material) and can play loud without clipping or sounding strained. It is also surprisingly open sounding and unstressed.

I love this tiny amp, especially the way it plays piano and violin - two very tough tests. I hardly miss my big Class A amp.

I would strongly recommended others to try out this board, or similar boards, or even other class D chipamp boards (there are tons of them from ST Microelectronics, Intersil, Tripath, International Rectifier, Zetex or Texas Instruments, and most of them have very positive user feedbacks). They are cheap to buy, though that's hardly the reason I'm recommending them, because they can stand firmly on their own sonic merits. I, for one, will surely try more similar boards in future. I have identified a few promising boards, but the TPA3118 or TPA3116 are fine starting points. My expectations were low despite reading of numerous positive user experiences. I could say I had almost zero expectation bias, but I'm glad to report that it ticks almost all the audiophile check boxes. But its biggest trait is it is unerringly musical. So bury your old biases against class D and do give it a try.
 
TPA3118 listening impressions

So bury your old biases against class D and do give it a try.


Last month I bought for one of my friend..and i was bowled over by the performance, when it drove JBL Synthesis speakers comfortably.
Only drawback I could find is the distortion at high volume level. For 11k what do you expect it to do...those who are hunting amp within 50k , should definitely give a try to this-

FX-AUDIO Pure Digital Amplifier D802C Remote USB Optical Coaxial Bluetooth AUX | eBay
 
One thing I missed mentioning - the 3118 has a very deep sound stage, making it quite holographic. Soundstage width is also quite wide. Voices and instruments have very clean and clear separation. It's quite good at dehomogenizing the various strands of complex music.
 
One thing I missed mentioning - the 3118 has a very deep sound stage, making it quite holographic. Soundstage width is also quite wide. Voices and instruments have very clean and clear separation. It's quite good at dehomogenizing the various strands of complex music.

I did listen to the modded tpa3118 monos yesterday night and I absolutely agree with what Joshua had mentioned. In addition to that very balacned tone and no shortage of dynanmics even with western classical music

It was sounding not bad for the money paid in stock form but now it's sounding genuinely good
 
Is there any off the shelf product ?

Check out SMSL SA-60. I believe it's based on TPA3116 chip. There are also similar finished products based on other chips like TDA7492/7498, etc.

Also check out Amptastic amp based on the Tripath 2020 chip. It is very well reviewed and has been continuously improved for something like a decade.
 
Get the Wharfedale EVO 4.2 3-Way Standmount Speakers at a Special Offer Price.
Back
Top