In a previous blog I described the benefits and challenges to the implementation of co-design. To better understand the concept of co-design, let us consider the design of an RF receiver with acceptable performance that will not render the algorithms ineffective, and in particular, a side-looking Synthetic Aperture Radar (SAR) on an airplane that covers an area on the ground (Figure 1).
Figure 1. Airplane mounted SAR.
As the plane flies, its antenna beam sweeps the area. Linear frequency modulated radar pulses are continuously transmitted and return pulses are collected. From the returned pulses, an image is reconstructed. This reconstruction is enabled by signal processing algorithms. The returned signals are at microwave frequencies and, therefore, have to be properly down converted to the baseband before processing them. The RF receiver must perform this task without significantly degrading the signal quality.
Rather than having radar transmit completely in baseband, an algorithm is used to take a picture, take the raster scan of pixels, analyze at the luminosity, and put the scans out as a data stream. The reflections of signals coming back into the radar front-end are then processed. Next, the long stream is taken, chopped into pieces and put into a picture.
While evaluating the quality of the images produced, the RF amplifier, RF mixer, down converting Local Oscillator (LO), amplifier nonlinearity, LO phase noise, and mixer characteristics are all considered. In this case, the LO phase noise caused the RF receiver’s contrast to suffer slightly (Figure 2). Although difficult to pick up with the eye, the impact can be seen in numeric measurements.
On the other hand, a significant visible impact can be seen with the amplifier going into 1-dB compression. As it gets close to its compression point, a great deal of detail and sensitivity is lost and washed out.
Figure 3. The change in image quality as A to D is changed from 14 to 6 bits can be observed from left to right.
Loss of detail comes from other sources as well. For example, let’s examine an Analog to Digital Converter (ADC). When it is moved from IEEE double-precision floating point to 14 bit to 12 bit and then to 10 bit, banding starts and there is a loss of detail (Figure 3). By the time six bits is reached, significant amounts of detail in the images are lost. At 6 bits,
an Automatic Gain Control (AGC) may need to be added on the front end. Auto scaling can be performed to get it into the sweet spot of the ADC.
In conclusion, in SAR, there are many contributing factors that can cause loss of detail and image quality. In order to compensate for this, co-design of DSP and RF can be used to produce better results.
Want to learn more about this example? Check out Dr. Murthy Upmaka's IMS 2016 MicroApps presentation.