Audio spectrum analysis

Note: since this article was published several of the oscilloscopes mentioned have been replaced with upgraded versions. The 16-bit ADC-216 has been replaced by the PicoScope 4262. The PicoScope 3224 and 3424 8-bit scopes have been replaced by a range of models in the PicoScope 3000 Series. The PicoScope 320x Series of budget scopes has been replaced by the PicoScope 2000 series ultra-compact oscilloscopes.

Advantages of Pico products for audio spectrum analysis

Spectrum analyzers tend to fall into two categories: so-called ‘swept’ spectrum analyzers and FFT-based spectrum analyzers. Swept spectrum analyzers work by using one or more notch filters (or mixers) to measure the signal amplitude at a given frequency, and by changing (or sweeping) the frequency of this filter a plot of amplitude against frequency can be constructed. Swept spectrum analyzers still have their place in high-frequency spectrum analysis, but for audio work they have the disadvantage that the signal must be constant for the whole period of the sweep.

FFT-based spectrum analyzers work by digitizing the signal of interest using a analog-to-digital converter (ADC). The stored values are then processed using the Fast Fourier Transform (FFT) algorithm. The advantage of this method is that the spectrum of a one-off or short-duration event can be captured. For example, using PicoScope’s trigger capabilities it is possible to capture the spectrum of a single drumbeat.

Performing spectrum analysis requires a lot of calculations, with some FFT-based spectrum analyzers taking several seconds to update a trace. PicoScope uses an optimized, high-speed routine for spectrum analysis that results in ‘real time’ results. Even on a relatively modest computer such as a 33 MHz 486 PC the spectrum analyzer can still update many times a second.

Pico PC oscilloscopes suitable for audio spectrum analysis

Although most of the PicoScope PC Oscilloscope range can be used for audio spectrum analysis, the higher-resolution devices are most suited. For high-end professional testing the ADC-216 is hard to beat. For general purpose work the PicoScope 3224 and 3424 (USB oscilloscopes) and ADC-212 (parallel port oscilloscopes) are ideal. If cost is an issue, consider the PicoScope 320x series.

The two key specifications for a FFT analyzer are sampling rate and dynamic range. A spectrum analyzer will be able to display up to one half of the maximum sampling rate. To cover the entire 20 kHz audio band this calls for a sampling rate in excess of 40 kS/s. If you are interested in testing the frequency response of amplifiers you may wish to look well beyond the 20 kHz point so a higher sampling rate is required.

The dynamic range of the spectrum analyzer is the next most important consideration. Most oscilloscopes (whether PC-based or benchtop) have an 8-bit resolution (256 steps). This limits spectrum analysis to 48 dB of dynamic range (20 log 256) The PicoScope 320x series are 8-bit devices. Unusually for oscilloscopes, the ADC-212 and, PicoScope 3224 and 3424 are 12 bit devices (4096 steps) which gives a theoretical maximum of 72 dB of dynamic range. The ADC-212 through a combination of oversampling, digital filtering and software averaging can actually improve on this theoretical 72 dB. The ADC-216 with its 16-bit resolution (65536 steps) has close to 100 dB of dynamic range.

To put these figures in context a typical tape deck would have 40 to 50 dB of dynamic range, a quality power amplifier 70 to 80 dB and a top end CD player 80 to 90 dB. As you will see below not all CD players live up to this.

The specifications of these devices are summarized in the table below.

Unit Resolution Sampling rate Spectrum range Dynamic range
ADC-216 16 bits 333 kS/s 166 kHz > 95 dB
PicoScope 3224 12 bits 20 MS/s 10 MHz > 70 dB
PicoScope 3424 12 bits 20 MS/s 10 MHz > 70 dB
ADC-212/100 12 bits 100 MS/s 50 MHz > 80 dB
ADC-212/3 12 bits 3 MS/s 1500 kHz > 80 dB
PicoScope 3206 8 bits 200 MS/s 100 MHz > 50 dB
PicoScope 3205 8 bits 100 MS/s 50 MHz > 50 dB
PicoScope 3204 8 bits 50 MS/s 25 MHz > 50 dB

Using the ADC-216 to test CD players

To show the sort of performance you can expect with the ADC-216 spectrum analyzer we decided to test two CD players. We chose a ‘budget’ portable model and a high-quality unit from Quad. One channel of the ADC-216 was connected directly to the portable CD player. The PicoScope trace below shows a pure 1 kHz tone from a test CD. As expected the result is a sharp peak at 1 kHz. The second, third and fifth harmonics are clearly visible showing distortion caused by the CD player. The peaks around 18 kHz are caused by the switching power supply inside the CD player's mains adaptor. If the CD player is run on batteries this noise disappears.

Next we repeated the experiment with the Quad CD player. As expected the results were much improved, the 5th harmonic is the most significant, 96 dB down on the main signal. The window showing measurements and harmonics is a separate program that took data from PicoScope (using DDE) to automate audio measurements. This program is no longer required since these measurements along with many others are now built in to the main PicoScope program. You can access the measurements from the settings menu—see help file for further details.

Crosstalk is an important performance indicator that can easily be measured with a spectrum analyzer. We played a 10 kHz sinewave (–10 dB) on the right channel of each CD player in turn (measured with the ADC-216). Ideally no signal would be present in the right channel, on the portable CD player the crosstalk is visible 60 dB down on the signal on the left channel.

On the Quad CD player, the crosstalk is at least 90 dB down.

An ideal CD player should have a flat frequency response over the whole audio spectrum. The specifications of our portable CD player stated a 20 Hz to 20 kHz response within 3 dB. We tested this using a sinewave that sweeps from 0 to 20 kHz. Plotting such a frequency response is not possible with many FFT spectrum analyzers as they take a quick snapshot of the signal then take several seconds processing and displaying the results. The result tends to be that only one frequency peak gets captured during the sweep. PicoScope’s data collection and processing are optimized for speed—even on a relatively slow PC (33 MHz 386) the spectrum analyzer has a near-instantaneous ‘real time’ update rate. The sinewave used for our test takes about 30 seconds to sweep from 20 Hz to 20 kHz. In this time PicoScope performs 100s of FFTs rather than the 2 or 3 that most FFT spectrum analyzers can manage. To display the frequency response as a single line rather than a moving peak, we used PicoScope’s peak detect function as shown below. As you can see the –3 dB point is not the 20 kHz claimed by the data sheet, but is nearer 16 kHz.

When the test was repeated on the Quad CD player, the frequency response was almost flat to 20 kHz. It also exhibits a sharper drop off after 20 kHz.

Several customers have asked for our advice on what type of signal source is best for testing amplifiers. The problem is finding a signal generator or sinewave source with a low enough distortion figure. We have been particularly impressed with the Black Star LDO100 low-distortion oscillator. The trace below shows its output at 1 kHz when plugged directly into an ADC-216.