Wide Band Vs Narrow Band

A simple spectrum agile radio would be a OFDM base band section followed by a heterodyne or direct conversion RF front end. The spectrum agility is obtained by using some of the subcarriers (M out of N) in the OFDM. The sub-carriers to be used by a cognitive radio is determined by a Spectrum Manager which gets information about the primary users from the Spectrum Sensing Unit.

Things get complicated when we start thinking about increasing the range of sensing. If suppose we could sense a very wide range of frequencies (We will see how this is possible later), then what should be the architecture of the spectrum agile radio.

If we use the same radio as earlier, it would be require a wide band RF front end which is capable of handling a big baseband bandwidth. Let us take some numbers to understand the situation better.

The frequency range over which sensing is done is say 800 - 2400 MHz i.e. a total of 1600 MHz.

If the spectrum agility is provided only in the baseband (which is primarily a digital circuit or could even be software running on a DSP/microprocessor), then the baseband bandwith should be more than or equal to 1600 MHz.

The data converters needed (DAC at the transmitter and ADC at the receiver) should have a conversion speed of few Giga-samples/sec. In fact, it should be more than 3.2 GHz. Though such high speed data converters exist in literature, most of them have very low precision (around 4-6 bits) and are primarily flash based architectures. This means high power consumption and large area along with low SNR.

Secondly, the RF front end blocks i.e. mixer, PA and LNA (plus IF processing stage in case of heterodyne receiver) should also be wideband. The Q of these circuits would be of the order of 1 which is very difficult to achieve at 1600 MHz.

The carrier frequency that needs to be generated is fixed, 1600 MHz. Hence the LO generation is not a problem. (I donno much about the phase noise requirement. But I think it should be a relaxed specification as compared to the existing narrow band communication standards.)

Now let us look into the baseband part. If suppose the minimum size of a spectrum hole is 200 kHz, then the total number of holes in the 1600 MHz bandwidth is around 8000. The major block in OFDM is an FFT/IFFT processor. An 8192 point FFT processor would take large area and power when operating at 1600 MHz.

If we look closely at the problem in hand, we will realize that the actual bandwidth used by a SU is very small. It would be typically vary from 200 kHz (for a GSM type channel) upto 10 MHz (for a high data rate user). Secondly, it is highly unlikely that the 10 MHz bandwidth required by an SU is scattered over the whole 1600 MHz range. More often than not, we will find the required spectrum holes within a bandwidth of 100 MHz.

This means a narrow band RF frond end (upto 100 MHz) would be sufficient for the SU communication. In order to have adaptability over the whole 800-2400 MHz range, we need a tunable RF front end. Along with that the LO should also generate frequencies from 850-2350 MHz (i.e. roughly 160 steps). Another problem that arises is the Q of the tuning circuits. The Q increases as the carrier frequency moves from 800-2400 MHz.

The base band portion would have a bandwidth of 100MHz and a 256 point FFT processor which is feasible.