Practical Chip Design

A conversation with mixed-signal chip and ASIC vendor austriamicrosystems at Electronica last week stirred up some interesting discussion about active noise cancellation. The company's immediate interest is in cellular handsets, but the question extends far beyond that application.

In general, the problem of active noise cancellation has just a few phases. First, one has to establish an heuristic to define the difference between the desired signal and the noise. This sounds obvious, but it is not necessarily a trivial step. Second, one has to create an algorithm for separating the two. An important part of that process—important enough to be considered a separate phase—is ensuring that once the separation is performed there is enough signal left to meet the needs of the rest of the system.

We are starting to see a simple form of this process now in handsets. The heuristic in this case is simple: signal is defined as anything originating in the near field, and noise as anything coming from further away. Not a bad approximation for speech from one party who is holding the handset. Not great for multiple voices, unless the participants are on very friendly terms.

The algorithm can be elegantly simple. If you place a second microphone on the handset so that it will be significantly further from the user's mouth than the primary microphone, there should be a large difference in amplitude between the speech signals from the two microphones. But the distant noise, since it is approximately equidistant from both microphones, will show little difference in amplitude. So just subtracting one signal from the other, maybe with a little bandpass filtering and phase adjustment, attenuates the noise much more than it attenuates the speech.

Now here's where austriamicrosystems comes in. The company has already established a presence providing mixed-signal subsystem chips (SSoCs?) that gather together the various relatively high-voltage analog functions that have to go on in a cell phone. These include switching regulators, battery management, the audio signal paths for speech, music, or whatever, and the audio DACs and ADCs. It is a natural extension to add the noise cancellation function as an analog subtraction and filtering step to this chip. The austriamicrosystems people point out that it is considerably more efficient in energy terms, and just works better, to do this operation in the analog domain.

Now for that third phase of the process. The audio signal coming out of the chip is now the difference between two analog signals. So under difficult conditions, such as high ambient noise, this difference signal may be considerably smaller than either of the microphone preamp outputs. And yet downstream functions, especially if this is a smart phone with yet another attempt at spoken-command recognition, may require a great deal of real dynamic range from this little signal. So the demands on the analog circuitry can be much higher, in terms of dynamic range, SNR, and total distortion, than they might first appear.

To see where this is going, consider a different design example: a collection of several, or many, microphones being used as a phased array. Such arrangements are already under consideration for automotive use, for example, and could conceivably end up in handsets as well. In this case, the heuristic is quite different. Signal is defined as sound coming from a particular direction—or if you want to get really fancy, a particular set of directions for multiple parties. Noise is defined as sound coming from anywhere else. To achieve this discrimination, the noise-reduction circuitry has to take in signals from all the microphones, alter their amplitudes and phases to select a particular angle of incidence on the array, and add them up.