Adaptive Baseband - Variable Bandwidth

A baseband architecture for cognitive radio using OFDM has been discussed in my previous post. The block diagram is again shown below.










Along with the tunable RF front-end and narrow baseband of around 100MHz as discussed in earlier posts, we can also think of making the baseband adaptive to cater to the needs of SUs with different bit rate requirements.

Let us consider the case of variable baseband bandwidth as shown in figure below.

Here the spectrum that is scanned for holes is from 800 MHz to 2.4 GHz. This range could be divided into 16 subranges of 102.4 MHz or 128 subranges of 12.8 MHz. Other divisions are also possible. The numbers 102.4 MHz and 12.8 MHz indicate the baseband bandwidth. The carrier frequency of the RF front end can upconvert/downconvert the baseband signal into any of the subranges.

The motivation for variable baseband bandwidth comes from the POWER perspective. As the baseband bandwidth increases, the power consumed by the data converters ADC/DAC increases for maintaining a particular SNR. In some ADC/DAC, like sigma-delta converters, the resolution and sampling frequency could be exchanged for one another, i.e. for the same power consumption, sampling frequency can be increased by compromising on resolution. In Sigma-Delta converters where the Effective number of bits (ENOB) is proportional to the Oversampling ratio (OSR).

Hence, the lowest power mode is the narrowest baseband bandwidth. That raises a question of the lowest possible baseband bandwidth. The lower bound depends upon the availability of the spectrum holes and the bit rate requirement of the SU.

Let me pose an example scenario. In the range 800 - 2400 MHz, the smallest size of the hole that is available is 200 kHz. This is dictated by the GSM band at 850/1800/1900 MHz. At any given time, the holes could be contiguous or scattered. If the spectrum holes are scattered in the spectrum, then the minimum baseband bandwidth required depends on the frequency separation of the farthest holes.

Consider the case when the holes are contiguous. Let the bit rate needed by the SU is 300kbps. The number of holes needed depends upon the SNR available in each of the hole. Let us assume that the 2 holes are allocated to the SU. Since the holes are contiguous, the baseband need not be OFDM and the baseband bandwidth can be 400kHz.

Is this possible?

The answer is NO because if the baseband bandwidth is 400 kHz, the number of subranges goes up to 4000. This puts a limitation on the frequency synthesizer in the tunable RF front end. The frequency synthesizer should be capable of generating frequencies in steps of 400 kHz in a range of 1600 MHz. This is infeasible with the present state-of-the-art.

Thus, OFDM is needed in the baseband even if the spectrum holes are contiguous. The infeasibilty of the frequency synthesizer restricts the minimum baseband bandwidth.

Moreover, the power consumed by the frequency synthesizer increases as the frequency steps get smaller and smaller (What is the relation? I donno). So the two conflicting terms, i.e. the power in the baseband and power in the frequency synthesizer, together decides the minimum baseband bandwidth supported by the architecture.


Summary
Known Facts

1. Minimum size of the spectrum hole = 200 kHz
2. Range of scanned spectrum = ~800 MHz to ~2.4 GHz ~ 1600 MHz
3. Power consumed in baseband processing increases with baseband bandwidth
4. Power consumed in frequency synthesizer increases with reduction in PLL step size

Unanswered Questions

1. Exact relation for power consumed vs baseband bandwidth
2. Exact relation for power consumed vs PLL step size

Baseband

The baseband part of cognitive radio is generally Orthogonal Frequency Division Multiplexing (OFDM). The basic block diagram of OFDM is given below.

The input bit stream is read 'k' bits at a time and mapped into symbol constellation, i.e. a symbol stream of complex numbers. This stream is parallelized into M channels. These M channels are then mapped to form one OFDM symbol. This OFDM symbol has M data sub-carriers, P pilot sub-carriers and N-(M+P) zeros. The zeros are inserted at dc and at the edges. Zero at dc is to avoid corruption of data due to dc offsets and LO leakage. The pilot data is inserted to estimate the channel for equalization and frequency synchronization. The number and location of pilot carriers depends on the method used for channel estimation. The signal at the input of IFFT is the frequency domain representation of the signal that is sent over the channel. A typical signal is shown below.

The frequency domain signal is converted into time domain signal by the IFFT block. A cyclic prefix is added to the data stream which is then serialized.

At the receiver side the data is parallelized and the cyclic prefix is removed. The FFT block converts the signal into frequency domain. Equalization is performed in the frequency domain, which is nothing but division of each channel by the gain of the transfer function of the transmission medium for that particular frequency. This data is then serialized and symbol demodulation is performed to obtain the bit stream.

If we make a tunable narrowband RF front end then the baseband bandwidth F_B would be around 100 MHz. See the last post for details about the need for narrowband front end.

In the case of cognitive radio, each SU uses only some of the data carriers. The spectrum manager will provide the SU with the locations of these data carriers. All other locations (frequencies) will be filled with zeros. These will include frequencies where PU is present.

Issues
1. Adjacent Channel Interference: When a SU is putting data in a sub-carrier, the resulting spectrum is not confined in that frequency alone. There is small amount of power transmitted in all the adjacent sub-carriers.

The resultant baseband spectrum is actually a sinc function. This will cause interference to the PU (as well as other SU). The interference is significant only at the adjacent channels as the sinc function decays pretty fast. The tolerable interference to the adjacent channel has to be defined and that may vary from one PU band to another. So for each sub-carrier, the amount of power permissible has to be calculated.

Guard bands, windowing techniques, insertion of cancellation carriers etc. are possible solutions for reducing adjacent channel interference.

2. Pilot carrier positioning: In the case of normal OFDM system, pilot carriers are usually scattered uniformly among the sub-carriers and the channel estimation at data carriers is obtained by interpolation from the pilot carriers. In cognitive radio, since only few of the sub-carriers are used for data (rest are zeros), a method for finding the location of pilot carriers needs to be investigated.

One of the possible solutions is to distribute them among each cluster of spectrum holes, i.e. there should be at least some P% of sub-carriers should be pilot carriers in each cluster of holes being used.

Another solution is to use long training sequences for channel estimation. In this technique, all the sub-carriers (being used by an SU) are used to send NT number of training symbols which are used to estimate the channel in each of the sub-carriers. This technique is useful only in slow fading channels. In case of fast fading channels, prediction of channel estimation will have to be employed with the training symbols correcting the predicted estimate from time to time.

3. Interference from PU: The received baseband signal of the SU will contain data from the PUs and other SUs. An FFT of the signal is taken, which gives the frequency domain information of the signal. Only those sub-carriers where the transmitter has put the data is used as the output and the rest is ignored.
A question arises whether the PU will add noise to the SU. The other SUs will not be an issue as they will be orthogonal.

4. Synchronization: In normal OFDM systems, the transceiver and receiver need to be time and frequency synchronized to avoid errors in demodulation. In case of many SUs sending orthogonal signals in the same band, the carriers of all the SUs need to be synchonized.