Upload
lekhue
View
224
Download
0
Embed Size (px)
Citation preview
11
Spectrum Sensing Technique using Polyphase DFT Filter Bank
for Opportunistic Cognitive Radios
Minseok Kim Jun-ichi Naganawa Jun-ichi Takada
May 29, 2009TCSR@Niigata University
Tokyo Institute of Technology
2
Background Cognitive Radio (CR) offers to solve spectrum
underutilization problem Spectrum sensing is an essential task to realize CR The challenges of spectrum sensing:
Reliability & Efficiency Opportunistic use of inactive sub-bands (white spaces)
White spaces (Spectrum Holes)
Frequency
Noise level
Now used Now usedNow used Now used
3Spectrum Dynamic Spectrum usage is dynamically changed in time and
frequency domain simultaneously How to detect the white spaces for opportunistic
cognitive radios is the main concern Frequency scanning is not suitable as the sensing
measurements become stale Wideband sensing scheme for multiple bands will be
required for efficient use of spectrum dynamic
Wideband RF front-end with wideband tuning function Settling time for channel switching is necessary Each band cannot be detected at the same time
Time Domain Narrowband Sensing
4
sensing sensing sensing sensing sensing sensing
Time
Channel switching & reconfiguration
RF
Variable LO
NarrowbandADC
2 T
Time
Frequency Domain Wideband Sensing
5
Search multiple unused sub-bands (white spaces) within wide frequency band continuously
No frequency sweeping Simultaneous spectrum sensing
Wideband ADC is necessary to sample wider band Dynamic range problem ADC resolution How to estimate the power spectrum density (PSD)
efficiently?
RF
Fixed LO
Spectrum Estimation
Wideband ADC
Time
Periodogram Spectrum Estimator (PSE)
6
21
0)()()(
N
kknxkhfY
fkjeN
kh 21)(
Most basic non-parametric spectrum estimator Apply discrete Fourier transform
)( 1ffH
)(nx
)( 0fY
)( 1fY
)( 2fY
)( 1NfY
Bandpass filters bank
Modulated Rectangular Window Sinc response in freq. domain
2
)(DFT1)( kxN
fY
)( fH
)( 2ffH
)( 1 NffH
7Dynamic Range & Resolution
MN /
0 5 10 15 20 25 30-50
-40
-30
-20
-10
0
Frequency [MHz]
Pow
er [d
B]
0 5 10 15 20 25 30-50
-40
-30
-20
-10
0
Frequency [MHz]
Pow
er [d
B]
0 5 10 15 20 25 30-50
-40
-30
-20
-10
0
Frequency [MHz]
Pow
er [d
B]
Conventional periodogram spectral estimator (PSE) by DFT filter has large sidelobe. The significant leakage of spectral power among different bands.
Limit spectral dynamic range Sidelobes can be decreased at the cost of wider main lobe by using windowing. Reduction of leakage among different bands is traded at the cost of a lower resolution in frequency Filter bank provides best performance to minimize the leakage power (best spectral dynamic range) keeping the frequency resolution. Prototype filter can be optimally designed
DFT Hanning Example prototype filterof filter bank
Spectrum Sensing Problems in Multiple Sub-channels
White spaces (Spectrum Holes)
Frequency
1H
0H
8
Noise level
Now used Now used Now used Now used
Power leakage from neighboring subbands
False alarm
helps detection
decided to be present
decided to be present
M bandpass filter array where each bandpass filter is the frequency-shifted version of a low pass filter H(w) with impulse response of a prototype filter
Filter bank can provide efficient bandpass filter implementation with low spectrum power leakage (good false alarm performance)
9Poly-phase DFT Filter Bank
M2
Mm2
MM )1(2
th0 st1 thm th)1( M
)(H
nd2
Analysis filter bank Bandpass filters for each sub-band
Output spectra
Polyphase decomposition
Discrete Fourier transform
Poly-phase DFT Filter BankM
-Point ID
FT
-1z
-1z
-1z
)(0 kx
)(1 kx
)(2 kx
)(1 kxM
)(0 kY
)(1 kY
)(2 kY
)(1 kYM
0H
1H
2H
1MH
)(nxM-to-1
M-to-1
M-to-1
M-to-1
)(0 ky
)(1 ky
)(2 ky
)(1 kyM
10
Filter Bank Sensing 11
A/D M-point ID
FT Filter B
ank
RF IF
Local OSC
sf 2
Decisions
2
00 T
11 T
11 PUPU NN T
10 /HH
0Y
21Y
1MY
The composite signal of multiple PU (primary user) sub-channels is down-converted into IF frequency. Then, the passband signal is sampled by wideband ADC and fed into the filter bank processing block.
The original idea is the minimum complexity design where the user sub-channels correspond exactly to the sub-bands (M=(Npu+1)/2).
Energy detector is followed by filter bank output and hypothesis test is performed.
Sub-band can be thought equivalently as
Test statistics for m-th sub-band
Test statistics for m-th primary user with Sp sub-bands
Energy Detection 12
MNN
)( mffH )( fX
M-to-1mY
m
m
u
li
N
ki kY
1
0
2)(pT
M: # of sub-bands
0
2
H:2
nGmT
22~ N
Filter gain
422 ,~ npnpp GNSGNS NT
1 ppp luS
1
0
2)(N
ki kYmT
False alarm probability for p-th primary user
Threshold level is determined by
Detection probability for p-th primary user
Energy Detection (2) 13
2
2
FAnp
npp,p
GNS
GNSQP
2FA
1npp,pp GNSNSPQ
22
22FA
1
22
22
D
snp
spnp,p
snp
snpp,p
NS
NSNSPQQ
GNS
GNSQP
Signal is supposed to be Gaussian process as well
14
Wideband Sensing Framework Filter bank can provide efficient bandpass filter
implementation with low spectrum power leakage (good false alarm performance)
Multi-channel sensing architecture based on poly-phase DFT filter bank followed by energy detector with minimum complexity Sub-channels have the same bandwidth (single
resolution) Wideband RF frontend license free radio using DTV bands in US(76-88, 174~216, 470~608, 614~698MHz)
…3836…1413…7 5165
76 - 88MHz 614 - 698MHz174 - 216MHz 470 - 608MHz
6 MHz DTV bands for spectrum sharing with unlicensed devices
15DTV bands for spectrum sharing with unlicensed devices
IEICE Trans. Commun. Vol. J91-B, No. 11, pp. 1351-1358
In case of DTV channel, the sampling rate at about 1.3GHz is necessary
Prototype filter :Energy in passband is maximized (Prolate sequence filter)
Wideband Sensing Framework 16
0 5 10 15 20-60
-50
-40
-30
-20
-10
0
Frequency [MHz]
Leve
l [dB
]
ISDB-T signalfilter bankPSE
PU1 PU2 PU3
RF signal BW : 18 MHz (=6MHz x 3) The composite RF signal with 3 PU sub-channels is down-converted into IF centered at 12MHz (=fs/4). Processing with passband signal(Real signal)
“Prolate spheroidal” or “Prolate” or “Slepian” sequence Real sequence of finite length and unit energy with the
minimum energy in the specified region Two parameters : Length L and
Prolate Sequence 17
2)( jeZ
1)(1 subject to
)(1 Maximize
0
2
0
2
deZ
deZ
j
j
1 subject to Maximize
1
1
ZZPZZ
Z is the eigenvector corresponding to max eigenvalue
18Simulation Specifications
Frequency
Noise level
PU 1 PU 2 PU 3
RF signal BW : 18 MHz (=6MHz x 3) The composite RF signal with 3 PU sub-channels is down-converted into IF centered at 12MHz (=fs/4). Processing with passband signal
19
Prob. Detection (Pd)
理想的(サブバンド分離が完全)にはどの手法も同一性能になる
PSEの隣接ユーザチャネルのノイズを拾ってしまうため,性能が劣化している(隣接ユーザチャネルは空いている場合)
フィルタバンクは理論性能に近い
-25 -20 -15 -10 -5 0 50
0.2
0.4
0.6
0.8
1
P D
SNR [dB]
FBPSEFB/PSE theoreticalSequential SimulatedSequential theoretical
Frequency
Noise level
PU 1 PU 2 PU 3
H1
001V
Frequency
(Proposed)
Pfa = 0.05
20
Prob. False Alarm(Pfa)
PSEの場合,隣接ユーザチャネルの信号電力を拾ってしまうため,隣接ユーザチャネル電力が大きくなるに連れて誤警報確率が高くなってしまう
フィルタバンクの場合,性能劣化無し
-25 -20 -15 -10 -5 0 50
0.2
0.4
0.6
0.8
1
P FA
Ps,neighbor/Pn [dB]
V=[001] FBV=[010] FBV=[011] FBV=[001] PSEV=[010] PSEV=[011] PSEPFA = 0.05 Frequency
Noise level
PU 1 PU 2 PU 3
H0
320 vvV
21
Prob. False Alarm(Pfa)(2)
PSEの分解能を向上し,隣接ユーザチャネルの影響を抑えることはある程度可能であるが,計算負荷が増加する
-25 -20 -15 -10 -5 0 50
0.2
0.4
0.6
0.8
1P FA
Ps,neighbor/Pn [dB]
V=[010] PSE (S=1)V=[010] PSE (S=4)V=[010] PSE (S=8)PFA = 0.05
Frequency
Noise level
PU 1 PU 2 PU 3 320 vvV
H0
3 4 5 60
2
4
6
8
10
log2(P)
Com
plex
ity R
atio
(CFB
/CPS
E)
S=1S=2S=4S=6S=8
22Complexity
P: # of sub-channels
S: # of sub-bands within sub-channels (PSE only)
S=1
S=2
S=4S=6S=8
23Summary Efficient wideband sensing scheme for multiple
sub-channels proposed By using poly-phase DFT filter bank, optimal
bandpass filter can be implemented Practical scenario (DTV in US) Future works
Optimum filter design Assessment in practical environment Dynamic range Optimum sensing strategy