Estimating power demand in mixed-signal ICs

NO PORTABLE PRODUCT IS LIKELY TO MAKE A MARKET IMPACT IF ITS BATTERY LIFE DISAPPOINTS BUT, EARLY IN THE DESIGN CYCLE, TRANSLATING SPECIFICATION-SHEET FIGURES INTO REALISTIC MILLIAMPERE AND MILLIWATT NUMBERS IS NOT STRAIGHTFORWARD.

BY ERIC HABER • WOLFSON MICROELECTRONICS

As designers of portable multimedia systems push battery life to the limit, they are spending more time than ever scrutinising powerconsumption data that silicon vendors provide. Like-for-like specification comparisons are usually difficult as there are too many variables, and crucial differences between competing parts are often far from obvious. Audio input and output subsystems are particularly tricky to compare as they include both analogue and digital circuitry and usually require several separate supply voltages. As a result, manufacturers’ data for such components is often irrelevant to real-use cases, and in some instances downright misleading. Armed with a basic knowledge of the relevant circuits, Ohm’s law, and a refusal to take manufacturers’ data at face value, the designer cansee through this confusing fog.

WHAT’S INCLUDED?

It may seem obvious, but understanding what circuitry each power-consumption figure includes is key to calculating a system’s overall power dissipation. However, with only a datasheet to go by, this may not be as simple as it appears. Consider the audio-output function in a portable system. Figure 1 shows the main blocks involved. The last few blocks in the chain—digital-signal enhancement, DAC (digital-to-analogue conversion), analogue mixing, and amplification— are usually integrated into a single component, which is loosely referred to as an “audio DAC”. However, when the datasheet for such a device specifies “DAC power consumption” or “DAC supply current”, it is very probable that this refers only to the DAC itself, excluding the amplifiers and other sub-circuits. But what if it says “playback to headphone”? Does that include on-chip signalenhancement functions like limiting, 3-D signal enhancement, or equalisation? Probably not, as silicon vendors can scarcely afford to make their parts look worse compared to their competitors. Some even specify DAC supply currents excluding the digital audio interface. Clearly, this bears no resemblance to any real-world use case, as the interface must be powered-up to receive audio data for playback.

To further complicate matters, system architectures also differ. For example, they might implement volume control in software on the CPU, or in the digital part of the audio chip, or using an analogue programmable-gain amplifier within the audio chip. A good sanity check is to identify what functions the architecture employs, then find out in which physical component they are located, and ensure thatthe power each function consumes is accounted for.

LOADING

Power dissipated within loudspeakers and headphones typically accounts for a large slice of overall power consumption. Since this power is not actually consumed within an IC, IC datasheets almost never include it. Fortunately, you can easily calculate it, as P=V_RMS²/Z, where V_RMS is the RMS voltage across the speaker and Z is its impedance (don’t forget to double that figure for stereo!). The hard part is choosing a realistic V_RMS. While you can readily calculate the largest possible V_RMS available from the output swing of the amplifier, in real life V_RMS depends on the end user’s volume setting. Even at maximum volume, it varies between louder and quieter passages within the same piece of music, so assuming a full-scale signal is overly pessimistic. Making a meaningful comparison between audio components requires a common benchmark. For example, the Japanese JEITA CP-2905B standard specifies that one should measure the battery life of systems with headphone output while driving 0.2mW (0.1mW per channel) into a 16Ω load.

WHAT’S THE SIGNAL?

The amplifiers that drive loudspeakers and headphonesare another particularly power-hungry type of component.

It is common practice for manufacturers to specify these devices’ power consumption in the “quiescent” state, i.e. playing absolute silence (represented in the digital domain as a long series of zeros). However, as soon as a real-life signal passes through the system, the power dissipated in the amplifier (as well as that in the load) increases. Clearly, amplifier supply currents should be specified with a non-zero signal—but what kind of signal? Manufacturers often use a 1kHz sine wave as it is easy to generate and some standards such as JEITA CP-2905B mandate it. However, this bears little resemblance to anything that real-world users listen to. Pink noise, as the IEC 60268-5 standard for loudspeakers defines it, is probably a better representation, although ultimately no one signal can reflect the infinite variety ofmusic.

When comparing amplifiers, it is also worth bearing in mind that their power efficiency depends on signal amplitude. The exact relationship depends on the individual amplifier (Figure 2). For example, under quiescent conditions, a Class D amplifier might well consume more power than an equivalent linear design, due to switching losses. Likewise, since linear amplifiers are more efficient at high volume than at low signal amplitude, their efficiency at fullscale can approach that of Class D amps. However, these extremes of signal amplitude are largely irrelevant—the battle for battery life is fought in the mid-range, where real-world amplifiers operate most of the time. This is where Class D amplifiers have rightfully earned a reputation as powersavers, beating comparable linear amplifiers by very largemargins indeed.

WHAT ABOUT THE DIGITS?

Amplifiers are not the only circuits whose power consumption is lower in the quiescent state than in real life. The same is true for other analogue circuits such as mixers and programmable gain amplifiers, as well as digital CMOS circuitry. In CMOS, power consumption is largely a function of how frequently bits toggle between the 0 and 1 states, so a signal consisting only of zeros (i.e. quiescent) results in unrealistically low supply currents. To get meaningful data,all components should be processing a real, non-zero signal.

Another factor to consider is the sampling rate of digital audio signals. Large sections of digital and mixed-signal circuitry switch once per sample, so their average power consumption is directly proportional to the number of samples per second. When comparing audio DACs or ADCs, you should try to find specifications for their supply currents that quote similar sampling rates. Further up the signal chain, the encoding quality of the source audio file (for example, the bit rate of an MP3 file) can affect the power consumption of the decoder. The bit rate, combined with the buffer size, determines how frequently data is retrieved from the storage medium. This is especially significant in hard-disk-based systems, where each disk access results in a large battery-current spike that can be a major factor in theoverall power consumption.

Many audio ICs, such as DACs or ADCs, can operate as either master or slave devices. In master mode, the audio IC drives the digital audio interface and therefore requires more current than in slave mode. It is hardly surprising, therefore, that power consumption is usually specified in slave mode. Does this mean that slave mode is always preferable? Of course not—after all, if the audio IC is not driving the interface, then the component on the other side has to do it, so the designer has merely moved power demand from one place to another around the system, not eliminated it. When you see a power figure that the manufacturer specifies in master mode, pay attention to the load capacitance, as this determines how much additional current the circuit needs. If datasheet figures assume large, “worst-case” load capacitances, the reality may be better than the specification. Conversely, IC vendors may “massage” power-consumptionfigures by using unrealistically low load capacitances.

Some audio components have special clocking modes that can eliminate the need for a power-hungry low-jitter PLL,which can only be used in master mode. For example, some Wolfso audio DACs and CODECs have a “USB mode”,where audio clocks are generated directly from a 12-MHzUSB clock. In this case, the power that the integration ofthe clocking saves usually far outweighs the power the audiointerface consumes.

POWER SUPPLIES

All but the simplest audio ICs use more than one supply rail. A typical circuit includes at least one analogue supply, a digital I/O supply for the audio and control interfaces, and a separate digital core supply. The IC’s overall power consumption is the sum of the power in each supply rail. An obvious way to save power is to use the lowest possible voltage for each supply. For the digital I/O voltage, the lower limit may be constrained by other system components to which the audio IC must interface. On the other hand, you can reduce the digital core voltage right down right to its lower limit, which datasheets normally list under “Recommendedoperating conditions.

Some datasheets include graphs of supply currents versus voltage in a given mode of operation. Where this is not the case, you can make some educated guesses. For digital logic in CMOS ICs, the current scales proportionally with voltage. This means that a voltage reduction is doubly beneficial— halving the supply voltage actually quarters the power that rail delivers. For analogue circuitry, things are more complicated because analogue circuits often contain constant-current sources. After halving an analogue supply voltage, the power that this part of the IC (excluding any load) consumes is generally somewhere between half and aquarter of its original value.

GETTING A CLEARER PICTURE

For a true, meaningful comparison of the power different audio ICs consume, the test conditions must be both realistic and consistent between the different ICs. This includes power delivered to the load, the nature of the signal (e.g. pink noise), sampling rates, and supply voltages. In addition, the functionality must reflect the desired use case; all required functions of the IC must be must be operating with all required functions enabled, and any functions not required disabled if possible. The digital interfaces of the audio ICs to be compared should either all run in master mode, or all in slave mode, and load capacitances should be the same in each case. Each IC’s master clock should also be the same; where a PLL supplies the audio clock, you shouldalso include its power consumption in the calculation.

In real life of course, different vendors tend to use differing test conditions for their audio ICs. However, knowing what factors most affect power consumption allows system designers to quickly spot omissions and to “extrapolate” from the vendor’s test conditions to their own real-world scenarios. This leads to a detailed view of IC power consumption, which is far more meaningful than the “headline” specifications you often find on datasheet frontpages.

AUTHOR’S BIOGRAPHY

Eric Haber is product architect at Wolfson Microelectronics.He first joined the company in 1999 and has held variouspositions, primarily in product marketing. He holds an M.Eng.degree in Electronics from the University of Edinburgh.

EDN Europe