Modified TA0104A Tripath Amplifier

I posted this thread last night (1/8/02) on another forum, but I thought this forum might lead to more discussion about this class of amp. I've added some links to make finding the parts easier.

I've now built 3 Tripath amplifiers over the last few weeks, the last being a no holds TA0104A ('104') version with a massive power supply (2-600VA transformers putting out +/- 82 VDC with 2 bridges and C-L-C filtering). The total cost to build it was a bit high: about $900, a little more than Chris Brodersen's version (http://people.mw.mediaone.net/cbrodersen/audio.html), but I'm also using two $100 input transformers.

The sound is simply OUTSTANDING, and beats many amplifiers that I've compared it to costing many thousands of dollars. This particular version that I'm describing is even better than the other two Tripaths that I built. Images are better delineated--there is a proper roundness, to say the resonances of a flute; depth is superior; and the soundstage is better proportioned; and dynamics/control improvements are palpable. This tweaked 104 produces 350W into 8 ohms and 770 watts into 4 ohms. (The other units either were the 103 with half the power, or used only one bridge rectifier and no chokes in the filtering supply.) My impression is that 2 bridges, one per power transformer, and choke filtering is superior.

There are several modifications that need to be done to the Tripath board as delivered in order to make it generate sufficient current into low impedance loads and provide stability. I found besides Chris' mods that the following is essential: paralleling a 4.75K resistor to R10 and R11, replacing C4 and C21 with a 250V polypropylene cap (the stock ones literally smoked and blew up!), the input caps (C13/16), add C6/7 using 33 microfarad/250 panasonic caps (EB series <- very important to avoid blowing up the cap; see Digikey at http://www.digikey.com/); parallel 1,000 pF cap at output jack (from Chris), but to this I paralleled a 470 ohm/5W resistor; replace R8/9 with lower values to get more gain (I used 20K with the 103 amp and 30K with the 104.

I did a few more tweaks that I'll describe below, but the above will give anyone an amazingly good amp. The Tripath board with heat sink is $300; figure, using round values, at least $100 for a chassis (http://www.par-metal.com/, I used the 14 series in a 2U height), $100 for a transformer (http://www.avellindberg.com/), $100 for regulated 5V supply (I built up my own PCB using an LM317) and filtering caps along with rectifier bridge(s). This totals to about $600. If you then add $100 for another transformer and filtering chokes, and $200 for input transformers--you get the $900 figure I mentioned earlier.

The input transformers are Jensen JT-11P-1 (http://www.jensentransformers.com/ln_in.html). I came up with this idea (and shared it with Chris) as a method to invert one leg and not the other in order to avoid power supply pumping, improving efficiency even more (this issue is described in a Tripath white paper, downloadable from their web site (http://www.tripath.com/audio.htm). (I've heard that some commercial versions use op-amps, but this means an inverted and non-inverted IC, leading to different sonic signatures between your channels.) The input transformers not only keep the sound the same, but also isloate noise and hum. These transformers do take away 1-2dB of gain, but I made up for this loss and more by replacing R8/9 above.

The stock Tripath amp inherently inverts the signal, but by inverting one input transformer and not that channel's output, that channel maintains absolute phase. The other channel is then inverted on the output, also maintaining absolute phase. BTW, these input transformers are used in Rowland preamps and amps and are very high quality. (As an aside, in this industry, the mark-up is about 5x over cost, so two $100 input transformers means $1000 more in the retail price of an amp.)

Except for adding a distortion LED output based on Pin 2 OVERLOAD output (this requires a P-ch MOSFET; I used IRF9610; too difficult to describe in this post), this covers my mods. If anyone decides to build one, good luck: it's worth the effort. This modified 104 amp seriously competes, and I believe betters, those costing 10 to 50 thousand dollar apiece, whose names I won't mention.

-Robert
 
Thanks for making the time to make such a detailed post! I've been experimenting with Tripath modules also, but have been side tracked to some other projects to complete first. Your post is very informative and helpful. I've used Jensen transformers in several projects, including an electronic crossover, and they are the best in their field for a readily available product.

Best regards,

Jon
 
I left the inductors the same as it seems fairly optimized for my loads. The inductors are the correct value and can handle the current. I suppose an outboard air-coil might be better, but I did not try this. However, I did change the 0.22 microfarads caps (C4/21) which are part of that same output filter due to the poor quality of the stock caps (as I mentioned one blew up).

As for the 12V regulated, I'm not convinced by examining the circuitry that it will make much of a sonic difference, but what do I know? What exactly did you notice with a 12V supply change?
 
Absolutely no RFI problems. I have the amp on the bottom of my rack; about 3 shelves up is my Dynalab Etude tuner, running as quietly as ever: no RFI.

This is in part due to steps taken to avoid them. First, 1,000 pF caps on the output terminals, as discussed on Chris' website, were used. Secondly, I use input transformers that block RFI. Thirdly, not previously mentioned, I use an input filter on the AC input jack (it's one of those large boxy devices that I ordered from Mouser that incorporates caps, inductors, and a line fuse). I can dig up the item number if anyone is interested. Fourthly, the C-L-C filter further attenuates high frequency garbage.
 
A few questions, if you don't mind

Robert,
Sorry to bother you with (maybe) trivial questions, but I don't know a lot about electronics and this would be my first "real" DIY project.
1) When you say "...2-600VA transformers putting out +/- 82 VDC with 2 bridges and C-L-C filtering...", how did you connect the two transformers? Also, what is a C-L-C filtering? Can you teach me how to build one or can you point me to some schematic? The ideal would be to get the power supply scheme of you amp.

2) What function do C6/C7 perform? On the original board from Tripath they are not stuffed. Talking about "not stuffed" components, there's a lot of them not stuffed, including a few transistors. Am I right in assuming it would be some sort of protection circuit? The documentation of the board does show a schematic for it, although the values of components are missing.

3) The capacitor and the resistor you put in parallel to the ouput do they run from the the "+" terminal to the "-" terminal or to the chassis ground? If I understand it correctly, the "-" terminal for the speaker should not be connected to the chassis...

I know this is a lot of questions, but I'm really a beginner and any help you could give me would be highly appreciated! Any useful hints or tricks are more than welcome.
Thanks
 
m.parigi,

Let me answer your questions in reverse order:

3) The 1,000 pF cap and 470 ohm resistor are both in parallel with the + and - output terminals as you inferred. You are correct in that the - terminal does not connect to the chassis; both outputs are 'floated' and are only connected to the 104 board at the indicated lcoations.

BTW, all references to parts are from the schematic, as I indicated in my original post, from the Tripath web site.

2) C6/7 are used to reduce undesireable current flows and 'clean up' the outputs. The required caps are relatively expensive, not essential for use, and are left off the demo board. However, since the space and markings for where to put them are on the board, it is a fairly trivial matter to solder them in. (You do need to open up the holes as they come filled with solder. Not to get bogged down in details, but my friend Julius Siksnius, who started Audire many years ago, taught me to use a toothpick to open up such holes. You heat the filled pad with your soldering iron and using a toothpick, poke clear the solder from the hole. Since these boards are double sided, you need to go back and forth between the sides to open up the holes. Don't overheat: it doesn't take much.)

As for the lack of components, most of what is missing on the 104 board is the output relay section. These parts are present on the 103 and other lower powered versions of this board. The reason they were left off, is due to the much greater current generated by the 104 version: it would fry the stock relay, so therefore the parts to control the relay were not used. The location where the relay would normally be is jumpered on the 104 board (I recall this is described in the documentation for the board). Alternatively, you could rig up a more powerful relay to handle the current if you want greater speaker protection. I elected to not use a relay on the 104 (but I left it in on the 103 versions that I've built).

1) This question is a little bit more involved than the others. I'll try to answer it, but you may need to search this forum for more discussions on power supplies and chokes.

Firstly, the transformers are each rated at +/- 30 VAC secondaries with center taps. Using only one would work, but I wanted a greater current capacity, so I used two: one for each leg (+/-). In this situation, the center taps are not used, so I trimmed that lead short and heat shrinked it to insulate it. Next, I paralleled each of the two secondary sections on each transformer. This meant I now had two transformers that produced 30 VAC with twice the current capacity and no center taps.

Next, each transformer was connected to its own rectifier bridge. From there, the bridge that is the positive supply had its negative output connected to the positive output of the negative bridge. These joined output is now your ground (center tap if you will). The positive bridge had its positive terminal then connected to the positive terminal/lead of the positive capacitor (described below) for the initial filtering. The negative bridge had its negative terminal next connected to the negative terminal/lead of the negative capacitor. (Remember that the negative lead of the positive filtering capacitor and the positive lead of the negative filtering capacitor both go to ground and each other.)

Secondly, the C-L-C refers to a power supply where the rectified DC voltage goes from the rectifier diodes to a capacitor filtering bank (one or more caps in parallel from either + or - to ground), then through an inductor (choke) in series with the + and - supplies (2 chokes are required: one for each rail), then onto another capacitor bank.

In my design I used 12,000 microfarads on the + rail and another one on the - rail, each rated at 100 VDC. The chokes were 12 mH each. Then another pair of 12,000 microfarads/100 VDC were used as the final filtering caps. From there the +/- 82 VDC (or so) went through fuses and then to the 104 board.

Aside from the above description, which I hope helps you, you probably need to search the internet for examples. One excellent resource is the Pass Labs site (http://www.passlabs.com/articles.htm), where you can download examples of how Nelson Pass designed not only his power supplies, but his amazingly good amps.

Good luck,

Robert
 
Thanks

Thanks Robert!
Your reply was extremely useful and clear.
I just have one doubt: is there a chance that you made a typo when explaining the transfomer wiring?

You stated "...I now had two transformers that produced 30 VAC with twice the current capacity and no center taps..."
If I got it correctly, it should be 60 VAC... otherwise I'm even more dumber that I originally tought!

Am I right? Thanks
 
You are correct, and I mis-typed (after, of course, mis-thinking). Thanks for pointing out this gross error.

In my paragraph following the '1)', that starts with 'Firstly,' I say the secondaries are paralleled on each transformer. Actually, I put them in series, as you point out. This series connection gives the required 60 VAC per transformer, which, once rectified gives over 80 VDC per rail.

And let me clarify the center tap matter again. There are two of them, one for each secondary (two per transformer), which should be insulated from the world. They are NOT connected to one another, or for that matter anything else.

I believe the rest that I wrote is correct for the moment...
 
Thanks for a great post, I picked up a lot of useful information.

Just a few questions in regards to your use of input transformers. The reasons I would like to use them are: a) to reduce EMI; and b) eliminate the electrolytic input capacitor. I was hoping to tie the center tap of the secondary to the 2.5V bias, and use the natural rolloff of the transformer as the input filter. Did you use this method, or are you just using the transformer for it's RFI filtering effect?

Any idea what the main purpose of the highpass filter is?
 
The input transformer that I'm using does not have a center tap. I don't know if you can bias around a center tap; maybe others could respond here.

The 1 microfarad cap used on the input (C13/16) of the 103/104 boards is essential in order to isolate the DC bias voltage from the load. Even with the input transformer, I could think of no way to avoid this blocking cap since the input transformer secondary would appear as a short circuit. I did replace it with a better quality capacitor, film capacitor (stock is a small electrolytic).

BTW, the input impedance is R8/9 + 5K. So, to get a little more technical than my initial post, I paralleled the secondary of the input transformer with a 15K resistor on each channel. This smooths out the load to avoid/minimize ringing on the secodary side. The Jensen JT-11P-1 ideally should see a 10K load. So if you use this transformer, and you vary R8/9, then calculate what value of resistor you need to approximate a 10K impedance. That is, if the final value of R8/9 is 30K (say a 25K resistor with the added 5K per the Tripath data sheet = 30K), then put a 15K in parallel with the secondary and it'll be looking at a 10K load.

The intended use of the input transformer was not to necessarily roll off the input highs, although it does serve that purpose at around 100 KHz. The main reasons I used them was to allow me to easily change input polarity, for the power supply pumping issue, and reduce EMI/hum/etc. I've found over the last few years that these transformers really make the background quiet.

(I have a friend, very respected in the Hollywood recording industry, who says many of them--the recording engineers--are migrating to using input and output transformers again, not only for the reduction of ground loops and hum, but for the quality of the sound.)
 
I'm getting close to having all the materials I need for starting this project. However, I'm having troubles in locating the 12 mH chokes in Italy.

Where did you get yours from? Are they air-wound or do they have a solid core?

I assume you shopped for at least 7 amps rating, am I right?

If you had them custom made, I'll try to get them replicated over here based on your input...
Thanks!
 
Robert,
Thanks for sharing your development work. The trannies are a wonderful method of accomplishing the phase inversion, as well of offering those intrisic CMMR (whether using single ended or balanced inputs) benefits available. A lower cost method (I'm certainly not saying better) is also available, the ti fully differential amps -
http://focus.ti.com/docs/search/paramsearch.jhtml?familyId=571&tfsection=param_table&templateId=3

If you look at the pdf file describing the THS4131 or THS4141, one can see that the circut is fully symmetrical and well as offering a really trick method of raising the input impedences of common mode signals, similar to the techniqe developed by Jensen and lisenced to THAT Corp. Additionally, the common mode reference point could be used to set the input voltage offset required by the tripath modules.

I've not built or heard circutry based on these amps, so I can't vouch for their audio qualities.
 
Getting there...speaker protection?

Robert,
I need some help with the @#%* chokes! Apparently all the people I asked for them in Italy had never heard of C-L-C filtering, hence their complete inability (unwillingness?) to manufacture the bloody objects.
Bear with me, but can you please tell me the details (i.e. part #, supplier, etc.) of yours, so that I can buy them in the US and have them shipped over here? Thanks.

On a different subject, what about protection of the speakers should anything go wrong (i.e. if a MosFet blows)? Is there a risk that the substantial currents of the power supply could destroy the speakers?

I had originally planned to build Rod Elliot's DC protection and turn-on-delay project, but that would involve putting a relay on the speaker output path. By reading the AKSA thread on this board I've seen that other members consider that a no-no, damaging to some extent the quality of the signal.
By the same token I assume that fuses on the output path are to be ruled out, right?

So, what did you do to be comfortable with the amp not killing your speakers?
 
I'd go with a Hammond choke. On their web page (http://www.hammondmfg.com/195.htm), I'd use the 5 mH (195G10) or 10 mH (195J10) version. Their address and phone is on their site. These cost about $40 each (not including shipping, etc) and will fit within a 2U sized enclosure. Each easily handle 10 amps continuous.

I use nothing special for protecting my speakers; the signal goes from the 104 board to the output terminals.
 
Output resistor: will it blow?

Getting close to powering the thing up!

I examined for the n-th time the schematic and I came up with a question.
By putting a 470-ohm resistor in parallel to the output you have excluded the possibility for the amp to work without load even if for a second.
The moment you remove the speaker load (because you trip over the cable :D or because you are a dumb DIY neophyte) all of the amp current will be funneled through the resistor, thus destroying it.
Am I right? Is there a way to prevent this to happen?
 
I wouldn't worry too much about it under most circumstances. You can leave it out if you wish. I provided it to make sure a load was always present for the amp. (When I was biasing one of the '103 amps, one channel kept blowing the output filtering capacitor without a load, so by providing one--and also upping the rating on the cap as I described in an earlier post--I stopped this activity.)

If, however, you were driving the amp at full power and pulled your normal load, you would have about 80V over the 470 ohm resistor. Using V=IR and P=IV, this would mean about 14 watts being dropped over that resistor. If you used a 5W resistor as I do, it would become rather warm with continuous, long term driving at 80V. By increasing it to a 1K/5W, and running the amp without any other load at 80V, it'll only be dropping a little over 6 watts, which would be OK. Going much larger will not provide much of a load for the amp, and increasing the wattage just makes the resistor too bulky.

Again, you can leave it out; it's just frosting.
 
hey rljones

Before I start asking questions, I would really like to thank you for posting your thread. I knew about the Tripath parts, but did not consider building them until I read your excellent post :cool: .

Now for the questions: Are you using the transformer because you are operating in bridged mode, or you chose to solely based on avoid the supply pumping issue? If I recall the Tripath white paper correctly, the pumping issue can be solved by using a large enough resevoir caps. I am tempted, if I need input inversion, to try pmkap's suggestion about the differential TI op-amps. Very clever :cool: .

What made you go straight for the eval board as opposed to buying the 104s and building a PCB? I know it's more work, but it's definitely much cheaper to put the parts together yourself. Tripath has quoted me $25 per module, $60 for the PCB (best fab price I can find), plus a few bucks for the MOSFETs, plus a few bucks for the rest of the passives (power supply excluded). Well, ok, maybe not much cheaper, but like 1/2 the cost or so that seems worth it to me :D .

I know Tripath's app notes cover this, but has anyone tried the IRFB41N15D's? My design will probably <60v rails, so I'd rather go for the MOSFETs with slightly better switching numbers and a lower Vds.

One more question. In app note 15, Tripath reccomends a new tweak to improve THD and efficiency at the same time. It doesn't appear to me like this is used in the eval board. Any idea why? I think this idea would be quite the killer tweak...

Thanks in advance.