Knowledge is power

Consumers are looking for more and better functionality on their portable multimedia player (PMP) systems. System designers have responded by creating increasingly complex systems. The challenge is that— notwithstanding the increasing complexity of demands in terms of system performance—consumers still expect continuous improvements in battery life.

This means that power consumption is crucial to designers, and they will spend many hours scrutinising the claims made by different silicon vendors. Their task is not made any easier by the fact that there are many variables in play, and in many cases manufacturers do not provide like-for-like comparisons between products. Indeed, the area of audio input and output subsystems is especially tricky, since they include both analogue and digital circuitry, typically with several separate supply voltages.

Reading into data sheets

A closer inspection of the circuitry involved in audio subsystems helps in understanding the true meaning of power dissipation figures in manufacturers’ data sheets. Figure 1 illustrates the main functional blocks involved in the creation of audio output for a portable system. Normally, the last few blocks in this chain – Digital Signal Enhancement, Digital-to-Analogue Conversion (DAC) and analogue mixing and amplification – are integrated into a single component, referred to as an “Audio DAC”. Where data sheets specify “DAC power consumption”, or, “DAC supply current”, it’s important to be sure whether or not this includes the power requirements of amplifiers and other sub-circuits associated with the DAC – if not, these need to be accounted for separately.

Likewise, quoted power consumption on data sheets for “playback to headphones” usually does not include on-chip enhancements such as limiting, 3-D signal enhancement and equalisation, all of which are frequently excluded from the power data by vendors to make their parts look better when compared with competitors. Some manufacturers even exclude the Digital Audio interface when specifying playback power consumption, which bears no resemblance to reality, as the interface must be powered up to receive audio data for playback.

Architecture choices

The fact that system architectures vary offers a further complication for designers. For instance, volume control may be affected via software on the CPU, in the digital part of the audio chip or using an analogue programmable-gain amplifier in the audio chip. As a rule of thumb, it’s a good idea to identify the relevant functions for the system being designed, ascertain which physical component holds that function, and make sure that the power consumption for each function is being properly accounted for.

Load and signal characteristics in the real world

There are other features of datasheets that often do not correspond to a real-world scenario. For instance, the power dissipated inside loudspeakers and headphones during playback accounts for a large percentage of overall power consumption, but these figures are not usually included in datasheets. More commonly, one will find power consumption specified in the “quiescent” (i.e. playing absolute silence) state, which is represented in the digital domain as a long series of zeros. In this state, voltage across the load is zero and no current flows through it. Whilst in a quiescent state, the audio IC itself consumes less power, which further reduces the headline power consumption – and sometimes, power consumption is even measured without connecting a load.

To get meaningful data, a load must be connected to the system. Typically in consumer electronics, its impedance is 8 Ohms for small loudspeakers and 16 or 32 Ohms for headphones. A realistic test signal must also be driven through all the relevant parts of the circuit and into the load.

Of course, the question of what constitutes a realistic test signal immediately arises. A 1kHz sine wave is easy to generate, which means that it is often used as a test signal – but this kind of signal does not reflect the mix of frequencies or the variations of amplitude over time that usually characterise music or speech. Perhaps the most useful kind of signal is provided by the IEC 60268-5 (formerly IEC 268-5) standard for loudspeakers. This standard uses so-called “pink” noise, which is a weighted mix of frequencies running right across the entire audio band. The “crest factor” – the difference between peak and long-term RMS amplitude – is well defined in “pink” noise, reflecting how real-world signals vary between louder and quieter passages.

Specifying signal amplitudes

Whatever test signal is used, its amplitude will dramatically affect power dissipation. This is yet another area where confusion can easily arise, since there are many different ways to specify signal amplitude. For example, “dBV” is relative to 1Vrms, whereas for “dBFS” the reference is “full-scale” – whatever that means for any given audio component. Decibel numbers using different references, or quoted without clearly specifying the reference, make it difficult to make meaningful comparisons. What ultimately matters is the power delivered to the load, so it makes sense to specify signal amplitude in terms of watts or milliwatts into a given load impedance.

Since the efficiency of any given amplifier varies with signal amplitude, it’s worth considering the power consumption of amplifiers across the signal’s entire dynamic range, as Figure Two shows. For instance, Class-G amplifiers use different supply voltages depending on the signal amplitude, and usually have a discontinuity around the switch-over points. In the example shown here, the amplifier’s Class-G mode saves around two milliwatts compared to a traditional Class-AB circuit at low amplitudes, whereas for louder signals there is no saving. But a new Wolfson development, dubbed “Class-W”, enables further savings compared to Class-G, and again the saving varies with amplitude, peaking at a signal level o 0.3mW into 30 Ohms. Very similar considerations apply for loudspeaker amplifiers, where Class-D technology is now considered the industry standard.

Sampling rates

Apart from amplifiers, there are other circuits in which power consumption is lower in the quiescent state than in real life. This situation also applies to analogue circuits such as mixers and programmable gain amplifiers as well as digital CMOS circuitry. In CMOS logic, power consumption is largely a function of how frequently bits toggle between the 0 and 1 states, so that a signal consisting only of zeros (i.e. quiescent) leads to an unrealistically low supply current. To deliver meaningful power data, all components should be processing a real, non-zero signal.

It’s also important to consider the various sampling rates in digital audio signals. Many parts of digital and mixed-signal circuitry switch once per sample, which means that their average power consumption is directly proportional to the number of samples per second. This in turn means that when designers are comparing different audio DACs or ADCs, the supply currents for each unit should be specified at the same sampling rate. Further up the signal chain, power consumption in decoders can be affected by the encoding quality of the source audio file – in other words, the bit rate of lossy file formats such as MP3. Combined with the buffer size, this bit rate will determine how frequently data is retrieved from the storage medium. This is particularly important in the design of hard-disk based systems, where each disk access causes a large spike in battery current.

Master versus slave modes

Audio ICs such as DACs or ADCs can be configured as either master or slave devices. This is important, because in “master” mode, the audio IC will drive the digital audio interface and therefore require more current than in slave mode. It will therefore come as no surprise that power consumption is usually specified in slave mode.

Of course, this shouldn’t be taken to imply that slave mode is always preferable – after all, if the audio IC isn’t driving the interface, then the component on the other side has to do it, so that the power requirement is simply shifted around the system, rather than being eliminated. A further tip: even if power requirements are specified in master mode, watch out for the load capacitance specification, as this will determine how much extra current will be required. If the datasheet figures assume large, “worst-case” load capacitances, then the reality may be better than the specification given. On the other hand, vendors may also be using unrealistically low load capacitances to bring the headline power consumption figures down.

Some audio components have special clocking modes that will eliminate the need for a power-hungry low-jitter PLL (phase locked loop). Many Wolfson audio DACs and CODECs, for instance, have a “USB mode” in which audio clocks are generated directly from a 12MHz USB clock. In this case, the power saved by integrating the clocking normally far exceeds the power consumed in the audio interface.

Low voltage power supply

Apart from the most basic ICs, all of these circuits require more than one supply rail. Typically, there will be at least one analogue supply, a digital I/O supply for the audio and control interfaces and a separate digital core supply. The overall power consumption for any IC is calculated by adding together the power (voltage multiplied by current) required in each supply rail – which means that the most obvious way to save power is to use the lowest possible voltage for each supply.

In the case of Digital I/O voltage, this may be given by the other system components with which the audio IC needs to interface. On the other hand, it’s possible to reduce the digital core voltage right down to its lower limit, which can normally be found under “Recommended Operating Conditions” in datasheets.

Ideally, datasheets would provide graphs of each supply current versus voltage in every possible scenario. Where such data is missing or incomplete, it’s possible to make some educated guesses. For instance, current scales proportionally to voltage in CMOS logic. This means that a voltage reduction is doubly beneficial – with a 50% reduction in supply voltage resulting in a reduction of 75% in the power used on that rail. In analogue circuitry, things are somewhat more complex, since analogue circuits often contain constant-current sources. Typically, though, after halving an analogue supply voltage, the power consumed by that part of the IC (excluding any load) is somewhere between half and a quarter of its original value.

Conclusion: The real story on power consumption

Summing up, test conditions must be realistic and consistent if the performance of different IC units is going to be accurately and meaningfully compared. Factors to bear in mind include the power delivered to the load, the nature of the signal (e.g. “pink” noise), sampling rates and supply voltages. In addition, the functions being compared must reflect the desired use case: all the functions required must be enabled, and those not required must be disabled where possible. The digital interfaces of the audio ICs being compared should all run in master mode, or in slave mode, and load capacitances should be the same in each case. The master clock for each IC should also be the same: where a PLL is required to derive the audio clock, its power consumption should also be included in the calculation.

Different vendors will use differing test conditions for their audio ICs. But if a designer can be aware of the factors outlined above that most affect power consumption, then this will allow them to spot omissions and to extrapolate data from the test conditions given to the real-world scenarios for which they are designing their systems. This will give designers a clearer picture of audio IC power consumption, a picture which is often very different and far more meaningful than the “headline” specifications found on the front pages of datasheets.


Captions

Figure 1: Example block diagram for audio playback, and factors affecting the power consumed in each block.

Figure 2: Efficiency of headphone amplifiers (example, using Wolfson’s WM8903 audio codec).

Click here for the illustrations:

Figure 1, Figure 2