TECHNOLOGY IN SYSTEMS
Analog to Digital Conversion
Compression-Based A-to-D Converters: Reaching New Low Power Limits in Quantization
The potential for power savings in analog to digital conversion can be significant if compression technology can be employed to reduce power consumption based on the signal’s characteristics.
FRED TZENG, ZEROWATT TECHNOLOGIES
Page 1 of 1
- Figure 1 - (a) Traditional quantization in the...
- Figure 2 - a) Traditional ADC power consumptio...
- Figure 3 - CDF of Rayleigh distributed signal.
- Figure 4 - Rayleigh distributed ultrasound sig...
- Figure 5 - ZeroWatt’s generation-1 compressi...
- Figure 6 - Industry ADC figures of merit versu...
- Table 1 - Performance results from the first ...
- Table 2 - Performance figures from the second...
In traditional analog-to-digital conversion, the signal is “blindly” converted without any consideration to the signal type or statistics. As a result, the conversion process wastes a great deal of power. If the signal type or statistics are used wisely, the analog-to-digital converter (ADC) power can be pushed to its lower limit. A new approach to ADCs uses a proprietary signal compression data conversion technique, Z-press, which delivers significantly lower power than traditional techniques. All these benefits come at no degradation to the signal-to-noise-and-distortion ratio (SINAD) and conversion rate compared to the traditional ADCs. Furthermore, Z-press needs no prior knowledge of the signal type. Simply plug in your signal input and the ADC will automatically adapt to the power-optimized state of the input signal.
The broader/commercial impacts of this ADC are major benefits in extended battery life, lower design costs, higher channel density and simplified system design in many electronic systems, particularly in medical imaging, wireless infrastructure, instrumentation and military applications. For example, in digital beamforming and phased array applications, there could be thousands of ADCs per device and therefore lowering power consumed by ADCs is a critical objective. With advanced ADCs that consume much less power, several critical objectives can be attained: 1) Battery life of portable devices can be almost doubled in certain systems, or 2) number of channels per device can be doubled, and/or 3) system resolution or speed can be enhanced, and/or 4) device form factor can be reduced due to higher channel density per chip. With these benefits, end users can gain more accurate detections with lower cost systems and enhanced portability.
Overview of Operation Principles
In traditional analog-to-digital conversion, the full voltage scale and input spectrum up to half the Nyquist rate (here defined as equivalent to the ADC sampling rate) is quantized, as pictured in Figure 1(a). As can be seen in Figure 1(a), all the red-outlined space represents area not instantaneously occupied by the signal in the amplitude domain. Quantization of this wasted space results in power inefficiency. In Figure 1(b) a signal compressor-decompressor engine is used around the ADC to focus only on the signal, resulting in little wasted space and significantly improved power efficiency. Similarly, the traditional ADC suffers from inefficient signal digitization in the frequency domain because the signal is not always occupying the entire spectrum band up to half the Nyquist rate, and in most applications is concentrated at certain frequency portions as shown in Figure 1(a). In contrast, the new approach can recognize that there is wasted frequency spectrum when not instantaneously occupied by the signal and quantize only the signal spectrum in a highly power efficient manner (Figure 1(b)).
(a) Traditional quantization in the amplitude and frequency domain, and (b) compression-based quantization in the amplitude and frequency domain.
Figure 2(a) shows the power profile of a traditional 10-bit, 50 MS/s successive approximation ADC digitizing the Rayleigh distributed ultrasound signal in Figure 3. This is compared to the power profile of a 10-bit 50 MS/s advanced compression ADC as shown in Figure 2(b). The traditional ADC is non-adaptive and uses no compression, while the use of compression data conversion enables the signal to be adapted to its lower power limit. Clearly the compression quantization method can achieve significant power savings, and depending on circumstances, can achieve more than 10 times compared to the traditional method.
a) Traditional ADC power consumption, and (b) ADC power consumption using compression.
CDF of Rayleigh distributed signal.
Compression ADC in Relation to Signal Probability Distribution Functions
One may wonder whether the compression ADC may benefit when dealing with real-world signals because both large and small voltage signals are present in a probability distribution. Furthermore, the same could be said about the frequency of the signal, as it may very well occur at half the Nyquist rate.
To consider how an advanced compression ADC can be very powerful under realistic signal conditions, we look at the cumulative distribution function (CDF) of a simple Rayleigh distributed signal shown in Figure 3, which is a signal that occurs frequently in wireless and medical imaging applications. In this case the signal is also backed off from the full scale by four times its average amplitude in order to avoid clipping of the signal. As can be seen, more than 85% of the time the signal is below 0.5V, so the compression ADC can work well in such cases.
CDF of Rayleigh distributed signal.
The benefits of the compression ADC can be better understood through a real-life application—ultrasound systems. In ultrasounds, the signal at the input of the ADC can be visualized by the signal in Figure 4. Multiple reflected ultrasound waves from the walls of the human body and organs arrive at the ADC with pulses of varying amplitude at arbitrary times. As can be observed, there are many “blank” spots between the pulses such that signal compression can offer a huge advantage. Traditional ADCs cannot simply turn off during the “blank” spots because they do not know when the next echo is coming back, and turn on/off time would range from 100s of nanoseconds to a few microseconds. Furthermore, the ultrasounds are usually pulsed carrier waves between 2-15 MHz while the ADC runs at 20-80 MS/s. As a result, the signal is oversampled 4-8 times at the ADC. Therefore, a great deal of the signal spectrum lies far below the Nyquist frequency, where again frequency compression can be advantageous.
Rayleigh distributed ultrasound signal.
Results of Compression ADC Designs and Future Work
ZeroWatt has been involved in low-power ADCs and has previously developed a first generation ultra-low-power ADC based on proprietary low-power compression techniques. Parts of this base design will be reused in the next round of adaptive compression-based ADC to lower the power even further. With the combination of first generation design and the new compression techniques, it is estimated we can bring more than 10 times lower power than the industry’s leading-edge ADC for certain signals. Measurements have been completed for the first generation compression ADC. A sample spurious free dynamic range (SFDR) measurement is shown in Figure 5. The results shown in Table 1 were accomplished in the first design.
ZeroWatt’s generation-1 compression ADC SFDR test with Fin=Fsample/7.1.
Performance results from the first generation design of the compression ADC.
The second generation adaptive compression ADC design uses a proprietary adaptive compression technique to lower the power even further. The adaptive method will allow the power consumption to wiggle dynamically according to the signal. Adaptation bandwidth and other parameters can be set by the user to optimize power consumption. The sampling speed, resolution and power achieved are 50 MS/s, 12 bits and < 5mW, respectively, suitable for digital beamforming and phased array channels. So far, the completed design is able to achieve the results shown in Table 2 for the next-generation ADC.
Performance figures from the second generation design of the compression ADC.
Technology and Competition
By using a compression and decompression system around our ADCs, we are able to bring the analog-to-digital conversion process to its lower power limit. Even with the best circuit design techniques, the technology is still far ahead in low power because we do data conversion based on instantaneous signal fluctuations. Figure 6 shows how the technology is able to outpace the typical industry average for ADC figure of merit (FoM), defined as
Industry ADC figures of merit versus compression ADC.
where ENOB is the effective number of bits and fclk is the sampling frequency. The plot shows various competitor products compared to that of ZeroWatt. The power consumption used here is based on a Rayleigh distributed signal, 8x oversampling, 4x full scale voltage back-off for the rms signal value, and implementation in a 0.13µm CMOS process.