IMHO it's not likely for 3x to deliver 250W with reasonable SQ and safety for several reasons. With 3x, each chip would see roughly an 8R load so, if you believe what Fig 18 says in the datasheet, you would have to use at least +/-40V on the rails to get to the 250W level. From what has been said so far in the thread, SQ suffers with rails that high. Pushing up the gain might help but it's by no means certain (has anyone tried?) and at +/-40V you may be pushing the ratings of some parts in the kit. Also, if you looks at what the datasheet says about thermal issues, sinking the heat could be a major challenge. You have to keep the case below 70C to dissipate 50W per chip and that is at absolute maximum ratings, among other thermal issues. See Table 2, Fig 17, Fig 18 and the accompanying text.
Last edited:
the load, effectively 8R at each chip, is well within ratings for the modular application. it's just when you try to get power up near 80+W per chip that things would get very rough on several fronts, most especially heat. so depending on your needs this could be a very usable design (or starting point).
Thanks! Much appreciated.
100KVR||39K=28k and all smaller figures.
(the 100k is a variable resistor)
At about 28k the midrange is soft. This can allow a little greater freedom in gain settings as well (without incurring as much shout if you attempt lower gain). At about 22k there's fairly balanced performance, to my ears and with my speakers and with my source device.
Indeed there is too many variables to predict, so you got an adjustable input load.
I did try to correct the text, but I probably didn't make it better.
Any suggestions?
The 3-chip, 135W won't run cooler than the 2-chip 90W, but there is a fair difference in output power as well.
TDA7293's are linear till 45 watts apiece, so the triple-parallel amp will do the same 135W with either a 4 ohm load or a 2.6 ohm load.
Oh, yes. With a 28+28vac transformer and 40x gain, one can do the extra effort of RC shunts across transformer secondaries and also using the filter at post30. One cost of high gain is the need of cleaner power. However, if cleaner power removes some noise voltage, then the gain can be set a little bit lower, perhaps 38x.sfthurber said:
And, yes, this can be used with the triple-chip amp and 4 ohm speakers since the wattage and voltage both arrive at the same 135W figure.
A 25+25vac transformer may be easier to use and allow slightly lower gain, and with the minor difference in voltage, the difference in decibels out of the speaker will not be audibly different (just a fraction of a decibel). When at full blast, transformer current (amperage) ratings will matter considerably more than a little voltage difference, because it is current that moves speakers.
P.S.
That 2.6 ohm speaker question: With 5 chips, (5*45) a linear 225W is possible, but the power transformer would be unfortunately expensive--if you use a triple load, you need a triple size (amperage) transformer. Instead of buying very huge transformers (well enough to blow the breakers if stereo), the funds might could be better used for more efficient professional 8 ohm speakers. 900va with 2.6 ohms. . . or a 4.5db more efficient 8 ohm speaker with 300va. . . both examples are equal in decibels.
Why the 2.6 ohm speaker? More information please?
P.P.S.
To avoid the on-chip limiter's bass reduction simply don't expose the "master" chip to more than 45W, and for example:
16 ohms 1 chip, linear to 45W (35W used)
12 ohms 1 chip, linear to 45W
8 ohms, 2 chips, linear to 90W (70W used)
6 ohms, 2 chips, linear to 90W
4 ohms, 3 chips, linear to 135W
These are the Watts Real Power figures, like a tube amp (that is all usable power if the transformer is strong enough); however, peak power may exceed these figures by more than twice, per the datasheet.
Last edited:
Here's an oddly effective enhancement:
I like a big long 1k, 15-turn, cermet trimmer, used as a variable resistor, dialed to 730R for the fb-shunt resistor, for two reasons. 1). adjustable and 2). reaches all the way from -in pin to coupling cap without going through any board traces. Those 1k 15-turn are available at the Radio Shack (and many other places)
And, I was able to dial it up to a slightly higher value. Yes, the chip likes it too and it tolerated slightly lower gain, 37X. So, it did slightly better imaging. Yay!
The ideal gain for a given chip amplifier varies mainly from power voltage, power decoupling and layout, and sometimes from RF/HF differences of a single component in areas that have open loop gain exposure, such as my big blue trimmer.
To do the same job that I did on accident (but on purpose instead), one could use ordinary parts and Cosine/Edmund style compensation (supplementary feedback involving capacitive divider with caps and/or RC's--these parts added to BOTH the fb resistor and the fb-shunt resistor), which will also result in slightly lower gain and slightly better imaging, without needing a big blue box stuck under the board.
As for me, I don't have the math skills to calc the Cosine/Edmund style compensation, so I'm keeping the big blue trimmer.
I like a big long 1k, 15-turn, cermet trimmer, used as a variable resistor, dialed to 730R for the fb-shunt resistor, for two reasons. 1). adjustable and 2). reaches all the way from -in pin to coupling cap without going through any board traces. Those 1k 15-turn are available at the Radio Shack (and many other places)
And, I was able to dial it up to a slightly higher value. Yes, the chip likes it too and it tolerated slightly lower gain, 37X. So, it did slightly better imaging. Yay!
The ideal gain for a given chip amplifier varies mainly from power voltage, power decoupling and layout, and sometimes from RF/HF differences of a single component in areas that have open loop gain exposure, such as my big blue trimmer.
To do the same job that I did on accident (but on purpose instead), one could use ordinary parts and Cosine/Edmund style compensation (supplementary feedback involving capacitive divider with caps and/or RC's--these parts added to BOTH the fb resistor and the fb-shunt resistor), which will also result in slightly lower gain and slightly better imaging, without needing a big blue box stuck under the board.
As for me, I don't have the math skills to calc the Cosine/Edmund style compensation, so I'm keeping the big blue trimmer.
Last edited:
as i think about it more, my suggestion still has one problem that the board itself has as well -- the "ground" of the FB-Cap is listening to a lot of ground current/noise that does not belong to its role. ideally i would think that the FB-Cap, like the FB-Resistor, should reference the sig output insofar as possible. see, e.g.: Neurochrome.com : : Audio : Taming the LM3886 Chip Amplifier at the section entitled "Improved Grounding Scheme" (ie: the FB-Resistor looks at the output point and the FB-Cap looks at the output's ground point.)
Last edited:
This may reduce the loop area.
But do look at the TWO loops. One has the flow current inside the chip, the other has the flow current outside the chip. The return current should be placed/located to minimise the loop area with it's respective flow current.
It would depend on where the Current Flows once inside the package.
Across the face of the chip may give a lower loop area.
It's my view that anything that reduces loop area gives a benefit to performance.
Sometimes that benefit is below the noise floor and thus difficult to show/prove.
Last edited:
yes and it's easy enough to try each way (at the chip face vs. below the board) but as you say the difference may get masked by other nasties.
i'm also inclined to try connecting the sigout line to the chip output pin and its return to the point between the PSU decoupling electrolytics on the board (cutting the board's long output traces out of the path).
i'm also inclined to try connecting the sigout line to the chip output pin and its return to the point between the PSU decoupling electrolytics on the board (cutting the board's long output traces out of the path).
2 more options:
It is fairly easy to groundlift (2R~12R) only the input (and not groundlift the fb-shunt).
Or, if we wanted to groundlift (2R~33R) to create a small signal star ground (groundpoint for both the input and the fb-shunt too), then we can simply bypass the groundlift resistor with a very small capacitor to prevent treble droop.
It is fairly easy to groundlift (2R~12R) only the input (and not groundlift the fb-shunt).
Or, if we wanted to groundlift (2R~33R) to create a small signal star ground (groundpoint for both the input and the fb-shunt too), then we can simply bypass the groundlift resistor with a very small capacitor to prevent treble droop.
Correct me if I misunderstand -- I read the last sentence of post 92 as implicitly acknowledging that there could be some legitimate applications for a ground lift, but if a ground lift is used then the sigin ground and the FB ground should not be on opposite ends of it. certainly you would not want any differential there to throw the FB loop off.
exactly.
The amplifier looks at the input and reads the DIFFERENCE between the Signal Flow and Signal Return. It processes that difference and outputs a signal.
You MUST NOT interfere with the amp's ability to read that difference signal. It needs that NFB to carry out the processing of that difference signal.
Daniel is on another planet.
i don't think there is any doubt about that. the question was how most accurately to derive that feedback voltage. here we derive the feedback voltage from a divider consisting of a series R and a shunt network (C and R in series). so there is the not so little matter of determining a suitable ground reference for the shunt. Andrew's conclusion was that it should be the same as the sigin ground reference. there are some, including a T.I design engineer, who have argued that it should be the sigout ground reference. see, e.g.:http://www.neurochrome.com/audio/?page_id=941
Last edited:
http://www.diyaudio.com/forums/chip-amps/252436-lm3886-pcb-vs-point-point-data.html#post3845470
In this thread Tomchr tries several options for the LM3886. Connecting Ri/Ci to the input ground with a short lead to the output/power ground gives the best THD results.
A good second method is to just connect all the returns to a single point. See post #30 for the conclusion.
The LM3886 datasheet also promotes this second method of grounding/returns.
In this thread Tomchr tries several options for the LM3886. Connecting Ri/Ci to the input ground with a short lead to the output/power ground gives the best THD results.
A good second method is to just connect all the returns to a single point. See post #30 for the conclusion.
The LM3886 datasheet also promotes this second method of grounding/returns.
The +input ground will tolerate a long trace, but the -in fb-shunt ground will not. The -in groundpoint has been DC filtered. After that, the -in groundpoint can't be exactly the same. That is on purpose and the dc blocking filtering is vital to good practical results.
However, it also means that the tolerances can't be found at centerpoint.
We would not like to consider that a surprise.
However, it also means that the tolerances can't be found at centerpoint.
We would not like to consider that a surprise.