TDA7293 single, bridge, parallel

[bnics]

Originally Posted by CharlieLaub

Anyway, my point is about the rising distortion above 1k Hz, not just at 20k Hz.

Since human ears are quite sensitive for under 3kHz frequencies, distortions above 1kHz are definitely problematic. Does anyone have graphs of distortions at say 2kHz or 5kHz?

Originally Posted by CharlieLaub

I am not sure where you get your info about power distribution - music does not follow a logarithmic power relationship, it's more like a band concentrated between 100Hz and 2kHz, with power dropping off on each end. This has been documented in published literature and on at least one DIY site I know of...

I had read about the logarithmic power somewhere. Since I can't recall or retrace references, err, well...

Could you post some pointers? Thanks.

[jcx]

some awareness of human auditory limits, music signal properties clearly is useful in trying to evaluate amp designs or performance from measurements

but I think we should also avoid "special pleading" - that some particular harmonic distortion or frequency distribution is "euphonic" and needn't worry us

Intermodulation is likely the bigger audible problem with nonlinear distortion - IMD points to reducing all nonlinear distortion mechanisms - including 2nd harmonic

with chip amps limited improvement can be had from paralleling to reduce loading effects, running inverting mode, Self's "XD" current bias of the output

to go further more feedback as in multiloop topologies and/or Black's Feedforward "Error Correction" would be the next level of performance improvement at a cost in complexity in design, compensation

the TDA7293 is problematic from a designers point of view with next to no datasheet information on GBW, phase margin that would let you more confidently design higher performance added feedback

[CharlieLaub]

Originally Posted by bnics

Since human ears are quite sensitive for under 3kHz frequencies, distortions above 1kHz are definitely problematic. Does anyone have graphs of distortions at say 2kHz or 5kHz?


I had read about the logarithmic power somewhere. Since I can't recall or retrace references, err, well...

Could you post some pointers? Thanks.

There is some limited info on music power distribution on this page:
Music and the Human Ear

I recently (last weekend) attended a presentation by Douglas Self at Burning amp in which he showed a nice plot of the "typical" power distribution in music. His point was that the power was falling off in the upper frequencies, above 2k Hz IIRC, to the extent that you could boost the signal in the active crossover high frequency section by 6dB or so and improve the S/N ratio by that much. He mentioned that he would post the presentation on his web site, but as of today I didn't see it. The plot also shows the power falling off below 100 Hz. In between 100 Hz and 2k Hz is a broad region of almost constant average power for music type signals.

-Charlie

[bnics]

Originally Posted by CharlieLaub

There is some limited info on music power distribution on this page:
Music and the Human Ear

Excellent article! A surprising quote from this article "It is also quite significant that even though the average tweeter power is low, the peak tweeter power is not all that much lower than other bands, and in fact is greater than the woofer in some cases!"

[AndrewT]

Originally Posted by CharlieLaub

between 100 Hz and 2k Hz is a broad region of almost constant average power for music type signals.

the average is regularly omitted from the statistics and from the thoughts of the readers.

Probably because it suits them and the argument they want to follow up with.

What have we got on data for transients peak signals over the frequency range?
Very little data, but I think it does not follow the shape of the average curves that most refer to.

My contention, due to an apparent lack of evidence, is that the peak transient signals have similar SPLs over the whole frequency range.

If this is true then it follows that the speaker driver/s should be capable of reproducing similar peak transient SPLs over the whole frequency range.

[tiefbassuebertr]

Originally Posted by panson_hk

I just made a test confirming that TDA7293-slave can be used solely as an output stage. This is done by not connecting the slave output and bootstrap pins to the master's pins. I still use the master buffer output to driver the standalone slave output stage. The diagram is shown below.

THD vs power for the master and the standalone slave is shown below. The slave-output stage is essentially operating in open-loop (no feedback) mode. Hence, we see very high distortion.

Regarding phase shift of the buffer output stage, I used a 200 kHz tone. Scope connected to master's driver output pin11 and the standalone slave output. The delay due to the buffer output stage is about 0.15 us.

Interesting observations.
BTW - the internal topology of TDA7293 output is very similar to that one about
http://www.amplimos.it/images/N-CH1.JPG
and
N-channel only output devices in a power follower.
go also to page 2/16 about the datasheet of TDA7294
http://www.datasheetcatalog.org/data...cs/mXqwvzw.pdf
The datasheet from TDA7293 don't show the internal topology.

Originally Posted by panson_hk

Hi Charlie,
We can find an opamp outperforming the front-end of TDA7293. The amp "opamp+TDA7293 output (slave)" may be a better amp.
I will try to see if TDA7293 can be operated in solely slave mode driven by another signal source. Panson

Interesting to know would be also the audible differences between the internal "front-end" of TDA7293 and an outdoor "front-end".

[sudhir81]

Some valuable information at ESP in this context

BiAmp (Bi-Amplification - Not Quite Magic, But Close) - Part 1

Relevant excerpt from this page -
X-over Frequency, Power to Bass (%), Power to Mid+High (%)
250 40 60
350 50 50
500 60 40
1,200 65 35
3,000 85 15
5,000 90 10

[AndrewT]

esp tells they are average power delivery in those frequency bands.
Average power is what helps a designer size the transformer and heatsinks.

For speaker driving ability it is peak or transient voltage in each of the frequency bands that matters.
For good sound quality the SPL peaks should not be clipped.

[sudhir81]

Thanks.

A hypothetical requirement: we want to design an amplifier for a single large 8Ohm full-range driver at 80W power. A quick lookup of LM3886 datasheet points to a +/-38V supply.

If we want to change it to a bi-amped setup with a 350Hz cross-over before two amplifiers - a smaller 8Ohm full-range driver and a large 8Ohm woofer. Average power division as per above table suggests 40W average power per driver. If we were to design a supply at +/- 27V, keeping the transformer VA rating same and decreasing secondary voltage, the amplifiers will run out of steam on peaks.

We may want to keep same +-/38V supply voltage and transformer secondary voltage to take care of the peaks. However since power dissipation will increase, so we'll have to increase transformer VA rating.

If some information about peak frequency distribution was available, we could design it at a lower supply voltage and only a little higher tranformer rating. Will that be a significant saving?

[AndrewT]

I doubt that there is a significant saving to be made.

I arrive at that tentative conclusion because of my belief that transient peaks can be of almost any frequency. Groups of transient peaks can be a combination of a narrow range of frequencies or a wide range of frequencies.
Considering the "music" as 20,000 samples per second then a 3minute track has some 3.6million music samples.
Not knowing much about statistics, I would venture to state that somewhere among those millions of samples there will be a large number of transient peaks. Maybe 10s, or 100s, or 1000s, or 10,000. It does not matter how many there are, what does matter is that there is a chance that some peaks consist of a narrow range of frequencies and some of a wide range of frequencies and the law of averages suggests that these frequencies can be anywhere in the audio spectrum, provided some operator has not deliberately moved the frequency response from flat to EQ.

Based on the above. The bass speaker may have to translate a bass heavy transient that approaches peak level and equally the mid/upper speaker may have to try to reproduce a predominantly treble peak.

Both halves of the "system" need to have peak SPL capability that is similar to the peak of the whole spectrum.

Now extend the listening experience from one track of three minutes to 1000s of hours of widely varied music types.
I can guarantee that somewhere in there, there are peaks covering virtually the whole range of frequencies in the audio spectrum.

Now extend your system from a two way to a ten way. Each driver having to reproduce only one octave in the audio spectrum. I think that statistics will tell us a different story. Since I'm no expert, I'll not even guess at the savings to be made in any single octave channel.