From SDR to Cognitive Radio

 
From SDR to Cognitive Radio
SDR has now been widely accepted as the implement of choice for a variety of platforms and
applications. The success in harnessing the promised flexibility and incredible processing power of
the SDR has led designers to consider implementing cognitive radios that adapt to their environment
by analyzing the RF environment and adjusting the spectrum use appropriately. The key components
for the successful implementation of cognitive radio are low latency and adaptability to the operating
conditions. These are the essential characteristic features that are needed for the deployment of
cognitive radios in multiservice scenarios such as communications, electronic warfare (EW), and
radar. Cognitive radios thus represent a huge evolution of SDRs.
Therefore, the cognitive radio has a lot to do with SDR [789–791]. As a matter of fact, the cognitive
radio works largely on the basis of many functionalities of SDR.3 It is of imperative importance
for us to understand how a software definable radio works in order to gain a better understanding of
cognitive radio. The discussion on SDR is to be covered in the subsection that follows.  9.4.1 How Does SDR Work?
An SDR is a collection of hardware and software technologies that enable reconfigurable system
architectures for wireless networks and user terminals. It provides an efficient and comparatively
inexpensive solution to the problem of building multimode, multiband, and multifunction wireless
devices that are able to work adaptively in a complex radio environment. In an SDR, all functions,
operation modes, and applications can be configured and reconfigured by various software. If the
configuration automation can be implemented in an SDR, a primitive cognitive radio will result.
The fundamental idea of SDR is to sample the received signal in the RF band right after the RF
low noise amplifier. It is also noted that the most important part of an SDR is its receiver part, rather
than its transmitter part. The reason is simple: the major difference between a conventional radio and
an SDR lies mainly in their methods of recovering required signals. Therefore, in this subsection we
will concentrate on the discussions on the SDR receiver.
The best way to describe what an SDR system looks like is to compare it with a traditional
heterodyne radio, as shown in Figure 9.4, which consists of a bandpass filter (BPF), a low noise
amplifier (LNA), a mixer, a frequency synthesizer, an intermediate frequency (IF) amplifier, an
automatic gain controller (AGC), a demodulator, an analog to digital converter (ADC), and a digital
signal processor (DSP), and so on. It is noted that filtering, amplification, and carrier down conversion
are implemented by analogue circuits. There might be several stages of IF amplification, thus needing
several IF filters, which makes it very difficult to miniaturize the terminal design due to their bulky
sizes.
On the contrary, in an ideal SDR receiver, as shown in Figure 9.5, the signal captured from a
wideband antenna will be directly sampled and analogue-to-digital converted; thus all postantenna
signal processing will be carried out in the digital domain. Therefore, the physical layer (PHY) air
interface signaling format will be determined “over the air,” or controlled by either a network or
a terminal operator. This feature is critical for the implementation of a cognitive radio. The only
difference is that a cognitive radio needs to scan a wide range of frequency spectra before deciding
which band to use, instead of a predefined one, as an SDR terminal does.
One of the most important characteristic features of an SDR terminal is that its signal is processed
almost completely in the digital domain, needing very little analogue circuit. This brings a tremendous
benefit to make the terminal very flexible (for a multimode terminal) and ultrasmall size with the
help of state-of-the-art microelectronics technology.
To implement an SDR receiver as shown in Figure 9.5, we have to raise the sampling frequency
up to at least twice as high as the carrier frequency seen from the antenna. For instance, if we are
interested in receiving the signals in a 10 GHz band, an ADC with a sampling rate of at least 20 Giga
samples per second has to be used. This will pose an even higher challenge if a cognitive radio needs to
scan an entire frequency spectrum up to millimeter bands. A compromise is to retain the RF front-end IF amplifier, and the signal will be sampled only at the IF bands, which will be much lower than RF
bands of interest and an ADC with a fixed sampling rate can be applied to all RF signals if the IF is
fixed. This can greatly simplify the architecture of an SDR receiver and lower the implementation cost.
An SDR receiver with IF sampling is shown in Figure 9.6, where different DSP chips will be used for
decoding different pay-loads carried in the RF signals.
9.4.2 Digital Down Converter (DDC)
An SDR terminal should be able to work under different air interface standards/modes. As mentioned
earlier, this requires that the signal be digitized as early as possible at a receiver, preferably right
after the antenna’s front end. However, the complexity of implementing direct RF sampling can be
formidable, so that the compromise that uses IF sampling is usually an attractive solution.
However, the use of the IF sampling technique gives rise to a new problem where the DSP
bandwidth and processing speed sometimes do not match the output signal from the ADC placed
after the IF amplifiers. Therefore, it is commonplace to use a digital down converter (DDC) to bridge
the gap between the DSP and the ADC output signal. The block diagram for the DDC is shown in
Figure 9.7, where signal processing algorithms can be explained by the following analysis.
First, the input wideband signal should be converted into complex baseband signal as
x[n] = r[n] e−j2πfcnTs = r[n] {cos (2πfcnTs ) − j sin (2πfcnTs )} (9.4)
where r[n] is the sampled IF signal, fc is the carrier frequency, and Ts is the sampling interval.
Now, this complex baseband signal is fed into an M-stage finite impulse response (FIR) filter, whose impulse response is h[m], to obtain
y[n] =
M
m=0
h[m] x[n − m] =
M
m=0
h[m] r[n − m] e−2πfc(n−m)Ts (9.5)
which can be rearranged to yield
y[n] = e−j2πfcnTs
M
m=0
h[m] r[n − m] ej2πfcmTs = e−j2πfcnTs
M
m=0
c[m] r[n − m] (9.6)
where c[m] = h[m] ej2πfcmTs is the combined coefficient of the FIR filter. Figure 9.8 shows the
frequency domain representation for the signals before and after DDC and filter-cum-decimation.
The use of the DDC in an SDR contributes to a great reduction of computation load in the
following DSP chip. In addition, the programmability of the DDC unit makes it possible to select
any portion of a signal as the one of interest. This brings about a great flexibility to an SDR. This
characteristic feature will also be very useful for the implementation of a cognitive radio.
9.4.3 Analog to Digital Converter
Another important element in an SDR is ADC, which performs the functions to sample, quantize
and encode continuous-time analog signals into a digital signal stream, suitable for digital signal
processing in the DSP unit. Obviously, the performance of an ADC unit will affect the overall
performance of the whole SDR system. In the following text, we would like to introduce several
important merit parameters for the ADC unit, which will be used in an SDR terminal.
The first merit parameter we want to discuss is the quantization noise. There are two fundamental
ways to perform quantization algorithms: uniform quantization and nonuniform quantization. The
nonuniform quantization algorithms include A-law quantization, μ-law quantization, adaptive quantization,
and differential quantization, and so on. In this book, we will only concern ourselves with
uniform quantization algorithm, whose quantization noise can be expressed by The signal-to-noise ratio (SNR) derived from quantization noise and aperture jitter can be
expressed respectively by
SNR = 6.02B + 1.76 + 10 log10
fs
2fmax  (dB) (9.8)
and
SNRaj = 20 log10
1
2πfmaxta  (dB) (9.9)
where B is the resolution of the ADC in terms of the number of bits, fs is the sampling frequency,
fmax is the highest frequency in the input analog signal, and ta is the aperture jitter of the ADC.
Other important parameters for an ADC are the signal to noise plus distortion (denoted by
SINAD), and the effective number of bits (denoted by ENOB), whose theoretical value can be expressed by
SINAD = 6.02B + 1.76 + 10 log
fs
2 · BW (9.10)
and
ENOB =
SINAD − 1.8
6.02 × B (9.11)
where B is the resolution of an ADC in terms of the number of bits, fs is the sampling frequency,
and BW is the signal bandwidth. More merit parameters for an ADC are listed in Table 9.1.
Obviously, the dynamic range of an ADC is proportional to its resolution in the number of
bits. Thus, the increase by one more bit in resolution results in a 6 dB increase in the dynamic
range. The second order intermodulation distortion (IMD) is generated due to the nonlinearity of the
ADC, producing an f1 ± f2 interfering component. On the other hand, the third order IMD produces
2f2 ± f1 and 2f1 ± f2 interfering components.
9.4.4 A Generic SDR
In this subsection, we will take a look at a generic SDR, which will be used to detect a DS/SS-BPSK
modulated signal. The block diagram for the SDR is shown in Figure 9.9, which consists of the
following major elements, a down converter, a BPF, an ADC and a DSP unit.
The transmitting DS/SS-BPSK modulated signal can be written as
s(t) = Ab(t)c(t) cos(2πfct + θ) (9.12)
where b(t) is a binary data information carried in the received signal, A is the signal amplitude, c(t)
is the spreading sequence, fc is the carrier frequency, and θ is the initial carrier phase. The received
signal can be expressed by
r(t) = s(t) + n(t) (9.13)
where n(t) is zero-mean additive white Gaussian noise (AWGN), whose double-sided PSD is N0
2 . As
shown in Figure 9.9, the received signal should first be fed into the mixer by multiplying a mixing
Table 9.1 Major merit parameters for analog to digital
converters for SDRs
A-D converter merit parameters
Sample rate
SNR (signal-to-noise ratio)
SINAD (signal-to-noise plus distortion)
ENOB (effective number of bits)
THD (total harmonic distortion)
IMD (intermodulation distortion)
SFDR (spurious free dynamic range)
Settling
Quantization noise
Resolution
Dynamic range
Differential nonlinearity
Integral nonlinearity
signal 2 cos 2πfLOt , whose frequency is fLO, where fLO < fc. The output can be written as
f (t) = r(t) × 2 cos2πfLOt
= Ab(t)c(t)cos 2π(fc + fLO)t + cos 2π(fc − fLO)t+ nm(t) (9.14)
where we have nm(t) = n(t) × 2 cos2πfLOt . The output from the mixer will be fed into an ideal BPF,
whose impulse response is h1(t). Removing the high frequency component Ab(t)c(t) cos 2π(fc +
fLO)t , we can obtain the IF signal as
m(t) = f (t) ⊗ h1(t)
= Ab(t)c(t) cos 2π(fIFt + θ) + nf (t) (9.15)
where we have fIF = fc − fLO and nf (t) = nm(t) ⊗ h1(t). m(t) will be sent into the ADC, using a
subsampling method to obtain
m(n) = m(iTs) =
∞
i=−∞
m(t)δ(t − iTs )
=
∞
i=−∞
m(iTs)δ(t − iTs ) + nc(iTs) + nq (iTs ) + nj (iTs) (9.16)
where nc(n) = nf (iTs ) is the zero-mean AWGN term after sampling, nq(n) is the quantization noise
generated in the sampling process, nj (n) is the noise term caused by the sampling clock jitter. It
can be shown that the combination of nc(n) + nq (n) + nj (n) is still a white noise. Performing the
Fourier transform against (9.16), we obtain
md(f ) =
1
Ts
∞
n=−∞
m(f − nfs ) (9.17)
which shows that the original signal spectrum will be reproduced in multiples of the sampling frequency
fs , as shown in Figure 9.10. If we let the sampling frequency equal the IF, we will obtain a
complete wanted signal spectrum at zero frequency, which can be sent into DSP for low-pass filtering
to remove other unwanted components before carrying out any other signal processing tasks in the
DSP unit.
It can be shown, under the assumption, that a Gaussian approximation method can be used for
performance analysis; the bit error rate (BER) performance for such a DS/SS-BPSK SDR receiver
can be written into
Pe = Q
√2 × SNR
where Q(z) is defined as and
Q(z) =  ∞
z
1
√2π
e−x2
2 dx (9.19)
and
SNR =
Eb
N0 + Nq + Nj
=
Eb
N0 + R2
3
 1
22Bq  1
fIF + 4R2
fIF #1 − exp(−2π2f 2
IFσ2
j )$
(9.20)
9.4.5 Three SDR Schemes
Based on the discussions given in the above subsection, we can obtain three major SDR schemes, as
shown in Figure 9.11.
Scheme A (as shown in Figure 9.11a) performs sampling in the IF signal, producing a complete
signal spectrum copy near zero frequency. Then it uses DDC to down convert it into a baseband to
perform the signal processing before the DSP.
Scheme B also performs undersampling in the IF band to generate a complete signal spectrum
copy at baseband, which will be sent into the DSP for signal processing.
The simplest scheme is Scheme C, which performs undersampling directly in the RF band, to
produce a complete signal spectrum copy at zero frequency, which is sent to the DSP for any other
signal processing tasks.
9.4.6 Implement Cognitive Radio Based on SDR
As discussed in the previous subsections, we can clearly see that SDR forms the basis for implementing
cognitive radio. The reason is simple: the SDR provides a very flexible radio platform, which can  be programmable and adaptively controlled by a central monitoring unit. The current state-of-theart
electronic technologies, including the ADC, the DDC, the high-speed frequency synthesizer, the
ultra-accurate timing controller, the microprocessor, the microelectronics fabrication process, and so
on, have made it possible to implement an SDR at a very reasonable cost and small size. The ready
availability of SDRs can make the implementation of a cognitive radio a reality in the near future,
although we fully understand that there are still many challenges awaiting us on the road to the
implementation of practical cognitive radios for commercial applications.
A conceptual block diagram of a cognitive radio scheme based on SDR modules has been shown
in Figure 9.12, in which there are two fundamentally important units, one being the receiver module,
as illustrated within the upper dashed line block, and the other the transmitter module, as shown in
the lower dashed line block in the figure.
Let us introduce the whole cognitive radio transceiver on a block by block basis as follows.
The first block on the left-side of the figure is the “wideband antenna,” which behaves like a
gate to the cognitive radio and controls the bandwidth the cognitive radio will operate in terms
of its RF frequency. As it has been acknowledged from the earlier discussions, a cognitive radio
may need to scan a fairly wide bandwidth to respond to the changing environment, and thus the
total bandwidth for a particular cognitive radio will depend on its applications and services. The
initial interest for the FCC in its NPRM [795] was the TV broadcast bands, which are not necessarily
used all the time. Therefore, the bandwidth of such a cognitive radio should cover all
those TV bands in the design of the wideband antenna, and so on. It has to be noted that the
total bandwidth for the “wideband antenna” is denoted by N
i=1
fi in the figure. This total bandwidth
should be divided into N sectors, each of which should be assigned to a particular SDR to
work on.
Also, a multiple antenna array is preferred for a cognitive radio, such that spatial beam-forming
(i.e., beam steering and null steering) and spatial diversity gain can be exploited to enhance the spatial
resolution and detection efficiency in performing various cognitive algorithms. As a matter of fact,
space, in addition to frequency and time, is another important domain a cognitive radio should take
into account to realize spectral reuse.
Following the “wideband antenna,” a “duplexer” will control the antenna sharing with receiving
and transmitting signals to provide a sufficient isolation between the incoming and outgoing signals.
The DFS, was originally introduced to avoid radar signals by IEEE 802.11a networks which operate
in the 5 GHz U-NII band, and here refers to an automatic frequency selection process, intended to
achieve some specific objective (such as avoiding harmful interference to a radio system with a higher
regulatory priority) in the cognitive radio.
There are N SDR units working in parallel in the receiver module of Figure 9.12, and each
will take care of a particular sector of the bandwidth of interest. The reason to use several parallel
SDRs, instead of only one single SDR, is that it needs to process a huge amount of data in each
bandwidth sector (
fi, where i = 1, . . . , N) before any kind of “intelligent” decision can be made in
the cognitive radio. On the other hand, we can also implement a cognitive radio using only a single
SDR unit. However, in this case, a very powerful SDR will be required to finish all data processing
within a reasonably short period of time. Unfortunately, it is still impossible to use currently available
DSP to fulfill such a challenging computation load for cognitive radio applications.
All output data will then be fed into a unit, which is responsible for making intelligent decisions
on them. Those decisions include the selection and combination of detected information, to obtain
the information we really want as the output.
On the other hand, the transmitter module in the cognitive radio scheme, as shown in Figure 9.12,
should carry out the tasks for information dispatch. The “adaptive synthesizer” works to generate the
correct local carrier reference to perform the modulation process and the up-conversion operation. To
do so, the transmitter module also needs useful information from the IPD unit, which provides the
current carrier frequency allocation chart, spectrum licencees’ program timetables and their transmitting
power profiles, and so on. This information is vital to determine a correct transmitting power level, so that the transmission from the cognitive radio will not interfere with existing incumbent
users. Similar functions will be performed in the “Timing Gate” unit, which controls the transmission
time slots, so that the transmissions from the cognitive radio will happen only when the spectrum
sector is free.
The generic layered architecture for the cognitive radio scheme shown in Figure 9.12 is illustrated
in Figure 9.13, where only two layers (PHY and data link layer) are shown because all the other
upper layers are applications dependent and thus not to our interest here.
The spectrum scanning is one of the most important PHY functions of a cognitive radio, scanning
all spectrum sectors of interest over the entire operating bandwidth. The scanning should also
be done to record the duty cycles of each carrier frequency, so that a cognitive radio will be able
to find the right time slot in the right carrier frequency to send the data. This will require the
ability to process a wide bandwidth of spectrum and then perform a wideband spectral, spatial,
and temporal analysis. It will be necessary for the cognitive radio to exchange their local sensing
information to optimally detect interfering incumbent users. This cooperation among different secondary
users in the same communication group will be important to estimate accurate interference
activities.
Channel measurement has to be used to determine the quality of scanned channels shared with
incumbent users. The channel parameters (such as transmit power, bit rate, and so on.) have to be
determined based on the channel measurement results.
A cognitive radio must have the ability to operate at variable data transmission rates, modulation
formats, different channel-coding schemes, and to transmit power levels. A multiple-in multiple-out (MIMO) system also might be used to suppress interference spatially and increase throughput via
multiplexing. The OFDM technique may also be used to further improve the bandwidth efficiency
and detection efficiency. The data transmission block will take care of all the above mentioned tasks.
There are other PHY functions, such as TPC and IPD, and so on, whose functions have been
discussed previously. On the other hand, the data link layer consists of three major blocks, including
“group management protocols,” “Medium Access Control (MAC) protocols” and “link management
protocols.”
It is assumed that any secondary user belongs to a secondary user group. The group management
protocols will be used to coordinate all secondary users in the same group. Any new user can get
all the necessary group information when joining a particular group. Link management protocols
take care of the link setup to enable communication between two secondary users and maintain the
link connection during the entire communication sessions. The data link protocol is used to select
a suitable channel to create the communication link. This selection should be made based on the
information obtained from spectrum scanning and IPD. Once the secondary link is established, the
data link protocols are responsible for maintaining the link connection. The MAC protocols work on
the basis of the information obtained from PHY, such as “spectrum scanning” and “IPD,” and so on.
MAC protocols will decide the ways of accessing the channels, depending on the patterns of shared
channels operated by the primary users.
The convergence sublayer in the data link layer is to provide a coordinating mechanism in a
cognitive radio to operate in different wireless environments, such as WWANs, WLANs, and WPANs,
and so on (also possible for WMANs, although it is not shown in Figure 9.13).
It is admitted that much work needs to be done before a complete layered architecture can be
designed. It should also be noted that the layered architecture has to be designed very carefully in
order not to increase the process latency, which is a very critical parameter in a cognitive radio. This
was the reason for suggestions that a cross-layer design approach [783, 784] may also be a choice
for cognitive radio. A primitive cross-layer optimization design for spectrum sensing is shown in
Figure 9.14.
We would like to use the following sentence to end this section. There are three primary signal
domains we can maneuver: time, frequency, and space, each of which has opportunities for spectral
reuse where a cognitive radio is concerned.