TI showed SemiAccurate a really interesting audio tech at CES called Smart Amp. Imagine applying the same rough ideas behind CPU/GPU ‘turbo’ to an audio amp and you have the rough idea.
Most amps today limit the power that can go to a speaker because it is not hard to damage the diaphragm through overexertion or overheat the coils/magnets if you push things. Spikes at certain frequency and power combinations will blow out your diaphragm and other combinations will overheat your coils so you have to lower power until everything is safe. This means speakers are capped at a small fraction of their potential maximum rating on the majority of frequencies to avoid the few damaging ones.
TI was claiming that a speaker that could take 10W for the majority of frequencies is usually capped at 2W output because of a few problem combinations that lead to physical damage. Not being an audio engineer, I can’t say if this is the norm or not, but the basic science behind it seems reasonable. Most people have heard a speaker clip or fail, there are certain patterns that lead to audible issues, the rest are fine.
If you recall the whole idea behind CPU turbo is that a device has a TDP, power supply circuitry has an upper bound, and each block of a chip has its own thermal limits. If you run everything at max speeds and capabilities, it will effectively burn itself to bits so there are conservative limits placed on the whole device. This problem leads to what is commonly called ‘dark silicon’. If you are aware of the power use and temperature of the whole device, you can tell when some things are underutilized and others would benefit from higher limits.
What turbo does is take unused power from one portion of a device and allow others to use it. It can also allow a block of transistors to run hotter than normal if other adjacent parts of the device are not putting out their allotted thermal energy. Lastly turbo can allow a block, core, or other portion of a chip to run at a higher than nominally rated frequency for very short periods of time until it hits certain thermal limits, then back down to a long-term safe limit.
Taken together this is why you see turbo ratings higher for one running core out of eight, a bit lower for two running cores, and almost nothing for all cores working hard. The chip knows what each part is capable of doing and what the whole is capable of, then limits individual parts on the fly to stay within those bounds. Problem areas are known and effectively mapped out in realtime.
That is where Smart Amp tech comes in to play. If you have a speaker that can take 10W for 95% of the frequency and power combinations but only 2W on those few problem areas, you need to cap it at 2W to be safe. If you could dynamically measure the sound patterns going to a speaker and limit those spikes, and only those spikes, to safe levels, you could run the speaker at 10W while only clipping the problems, not the whole waveform. The same goes for overheating the coils, if you know what the specific problem patterns are, dynamically limit them and only them.
Clip only problem areas, let the rest go
This of course means you need to know exactly what the problem areas for each individual speaker type are, not a trivial thing to discover. Luckily if you characterize the speaker during its design and testing process, you can figure out exactly what causes problems. TI has a test suite for this and essentially measures the overheating via electrical resistance while physical damage can be measured in an automated fashion too. In both cases you probably need a few examples to test to death, but the overall process isn’t too complex.
Once you know your problem areas, you can set the Smart Amp to analyze the incoming signal and clip only areas that it knows will damage your hardware. The rest of the signal can run at a large multiple of the normal conservative power limits that would otherwise need to be set to avoid damage, a clear win for almost all devices.
That brings up the problem of sound quality, if you are clipping certain frequency and power combinations or limiting other patterns for overheating, it does change the sound? How much does it change things? These are not easy questions to answer because it depends on how much you push the limits. In the 2W/10W example above, if you raise the power limit to 4W you probably won’t clip very much at all, 10W will clip much more. Sound quality becomes another variable that you trade-off in system design, a knob you can now turn that wasn’t there before.
During the demos SemiAccurate saw at CES, there was no noticeable change in sound quality between the ‘safe’ limit and the Smart Amp allowed limit on a sound bar type speaker. With the same physical speakers the Smart Amp version was noticeably louder but I personally could not hear any differences or problems. There isn’t any audible clipping at all with the caveat of this is a manufacturer demo. That said since the Smart Amp shouldn’t drop out those problematic combinations, just limit them and only them to safe levels, any audible effects should be quite minimal. Lets just say for me it worked just fine.
For the majority of speaker uses things are not space or power constrained and signal integrity, IE sound quality matters quite a bit. For these uses, we can’t see Smart Amp technology being worth a great deal, nice to have maybe but not worth the tradeoff in most cases. Do you want you high-end audio system to sound perfect or be louder? Is fidelity more important than decibels? Since it is a tradeoff, implementing Smart Amp on a system does mean you can have both but where the cutoff lies is an open question and quite possible a user controllable one.
When you start looking into space constrained use cases like phones, tablets, laptops, and other places where you can’t just use a bigger diaphragm, Smart Amp becomes a very useful tool. Would you trade-off clipping a few frequencies here or there for five times the power output through the same speakers on your phone? How about the speaker phone in your conference room? Bluetooth headset? Earbuds? Need I go on?
For use cases where there are physical constraints on the speaker, Smart Amp could be the proverbial killer app. In those cases you would gain vastly more from higher power throughput than you would lose from a bit of minor targeted clipping. Because of that I think TI’s Smart Amp tech has a very bright future ahead of it. Just like turbo has taken over in silicon, dynamic tuning of the signal to match the speakers will become an expected baseline feature in short order.S|A
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018