SIGNAL CHAIN BASICS #69: Deliberately increasing THD in audio processing systems: How and why?
For those of you trying to produce higher fidelity from your audio system, here’s a new concept for you. Many systems, especially those in the home theater / mini-micro combo market, deliberately add distortion to the output signal. While this may seem very counter intuitive, there are a few reasons why designers may consider doing it. The main function of this technology is to maximize the average power output, while limiting peaks.
Some customers use the same power amplifier ICs in a range of products. This affords them higher volume on one device, to lower costs, and simplify inventory. They may use a lower total capacity power supply to save cost. The customer may use a closed-loop, fixed-gain amplifier with a small power supply. This limits the output voltage swing (by limiting the output), which can protect a smaller power supply from over-current conditions. However, a simple attenuator makes the system sound much quieter. Distorting the output a little adds significant perceived RMS power. The careful balance is in deciding how much distortion is too much distortion!
For other customers, limiting the voltage output of their signal can help with limiting the excursion of the loudspeaker. Yet, care should be taken in this case as higher RMS power into the speaker may cause reliability issues of its own.
In a digital processing system, THD can be introduced to a signal by saturating the digital sample. That is, applying so much gain that the most significant bit is pushed beyond the size of the digital sample. For instance, you have a 24-bit word and your sample is 0x900000. By applying 12 dB of gain, the highest audio bit is pushed beyond the most significant bit (MSB) in the sample.
The trick then is to scale that data back down to the audio output level that you want. So, in summary:
Figure 1. Boosting a signal into clipping increases THD, then lowering output creates more average power for a given peak-to-peak voltage.
This may sound simple, but many audio processors don’t actually have their most significant bit = full scale audio. For example, some of TI’s audio processors use a data format known as 9.23. This sample data represents 16- or 24-bit data in the following way:
Figure 2. Mapping a standard 16- or 24-bit audio sample into 32-bit or 48-bit memory location.
As you can see, padded bits are added at the MSBs and LSBs. The LSBs are easily understood – if you attenuate a 16-bit word (from a CD player), then you still have bits that you can replicate without truncation.
At the upper end, nine bits are there to protect the audio data from accidentally saturating. For instance, if you have an equalizer (EQ) with a 24-dB boost and you input a “full scale” 16-bit word, you may unintentionally saturate the signal or add distortion, which is the opposite of what we’re trying to achieve.
There is amplitude loss when clipping, so THD (post) might allow slight gain through THD manager. Ten percent distortion clipping accounts for approximately –1dB of output level loss.
Let’s run through an example
In our example, the system has a 9.23 audio path. We want to create 10 percent of THD at –12 dB of the output. The average input is –10 dBFS (–10 dB reference to the 24-bit full scale audio source).
We need to boost to full scale and beyond (nine bits of “overflow bits”). So, in a boost function, we add 10 dB to original source to get to full scale, then another 27 dB to fill the nine overflow bits. Now add 3 dB of gain to clip the signal. In total, we need to add 40 dB of gain.
We now have a signal that has filled even the MSB of the audio path, and requires attenuating so that the output content is at –12 dB. That means attenuating by 39 dB. The resulting output has 10 percent distortion with output level –12 dB. Voila! We now have increased the RMS power (by increasing the distortion) at an output of –12 dB, and simultaneously made life easier for the power supply and speaker.
For those running with graphically programmable processors such as the PCM3070, this can be prototyped quickly and auditioned using TI’s PurePath Studio Graphical Development Environment.
For more information on PurePath, visit: www.ti.com/purepath-ca.
Please join us next month when we will discuss digital correction of analog quadrature modulator imbalances.
For previous SCBs click here.