GHz-Wide Sensing and Decoding Using the Sparse Fourier Transform
Summary (5 min read)
1 INTRODUCTION
- The rising popularity of wireless communication and the potential of a spectrum shortage have motivated the FCC to take steps towards releasing multiple bands for dynamic spectrum sharing [1] .
- Fig. 1 from the Microsoft Spectrum Observatory [2] shows that, even in urban areas, large swaths of the spectrum remain underutilized.
- In particular, realtime GHz sensing enables highly dynamic spectrum access, where secondary users can detect short submillisecond spectrum vacancies and leverage them, thereby increasing the overall spectrum efficiency [4] . .
- The authors explore how one can achieve the best of both worlds.
- Like the compressive-sensing approaches, BigBand can acquire a wideband signal without sampling it at the Nyquist rate.
3 BIGBAND
- BigBand is a receiver that can recover a sparse signal with sub-Nyquist sampling using low-power commodity radios.
- Thus, BigBand can do more than spectrum sensing -the action of detecting occupied bands.
- BigBand adapts the sparse FFT algorithm for spectrum acquisition using low speed ADCs.
- BigBand discovers the occupied frequency positions f and estimates their values x f .
- Once x is computed, it can recover the time signal x and decode the wireless symbols.
3.1 STEP 1: Frequency Bucketization
- BigBand starts by hashing the frequencies in the spectrum into buckets.
- BigBand then focuses on the non-empty buckets, and computes the values of the frequencies in those buckets in what the authors call the estimation step.
- Recall the following basic property of the Fourier transform: sub-sampling in the time domain causes aliasing in the frequency domain.
- Further, the value in each bucket is the sum of the values of only the frequencies that hash to the bucket as shown in Eq.
- It then focuses on the occupied buckets and ignores empty buckets.
3.2 STEP 2: Frequency Estimation
- Next, for each of the occupied buckets the authors want to identify which frequencies created the energy in these buckets, and what are the values of these frequencies.
- If the authors can do that, they then have recovered a complete representation of the frequencies with non-zero signal values, i.e., they acquired the full signal in the Fourier domain.
- Even for the occupied buckets, the sparsity of the spectrum means that many of these buckets will likely have a single non-zero frequency hashing into them, and only a small number will have a collision of multiple non-zero (or occupied) frequencies.
- To compute f , the authors leverage the phase-rotation property of the Fourier transform, which states that a shift in time domain translates into phase rotation in the frequency domain [38] .
- Alternatively, one can use different delay lines to connect the clocks to the two ADCs.
3.3.1 Collision Detection
- Again the authors use the phase rotation property of the Fourier transform to determine if a collision has occurred.
- Since the colliding frequencies rotate by different phases, the overall magnitude of the bucket will change.
- Thus, the authors can determine whether there is a collision or not by comparing the magnitudes of the buckets with and without the time-shift.
A. Resolving Collisions with Co-prime Aliasing Filters
- One approach to resolve collisions is to bucketize the spectrum multiple times using aliasing filters with co-prime sampling rates.
- Co-prime aliasing filters guarantee (by the Chinese remainder theorem) that any two frequencies that collide in one bucketization will not collide in the other bucketizations.
- The two frequencies labeled in red and blue collide in a bucket whereas the frequency labeled in green does not collide, as shown in the figure .
- The authors then go back to the first bucketization and subtract the blue frequency from the bucket where it collides to obtain the red frequency.
B. Resolving Collisions without Co-prime Aliasing Filters
- Co-prime aliasing filters are an efficient way to resolve collisions, but they are not necessary.
- This means that one can use one type of ADCs for building the whole system.
- Hence, the authors can solve overdetermined system for the possible (f , f ′ ) pairs and choose the pair that minimizes the mean square error.
- The authors empirical results (in §7.3) show however that for practical spectrum sparsity (which is about 5%) 3 shifted bucketizations are enough to uniquely resolve the colliding frequencies.
- One uses this method to resolve collisions of two frequencies while iterating between the co-prime filters.
4 CHANNEL ESTIMATION AND CALIBRATION
- The earlier description of BigBand assumes that the different ADCs can sample exactly the same signal at different time-4.
- This makes it possible to build BigBand using only USRPs [39] . shifts.
- To better understand this problem, let us consider the case where the authors resolve collisions without the co-prime sub-sampling.
- The hardware channels are different for the different bucketizations.
- The authors need to estimate them and compensate for them in order to perform frequency estimation and also resolve the collisions.
4.1 Estimating the Channels and Time-Shifts
- To estimate the channels and the time shifts, the authors divide the total bandwidth BW that BigBand captures into p consecutive chunks.
- Both the magnitude and phase of the hardware channel ratio will be different for different frequencies.
- As expected, the phase is linear across 900 MHz.
- These hardware channels and time shifts are stable.
- For their implementation, the authors estimated them only once at the set up time.
5 SENSING NON-SPARSE SPECTRUM
- The key idea is that although the spectrum might not be sparse, changes in spectrum usage are typically sparse, i.e., over short intervals, only a small percentage of the frequencies are freed up or become occupied.
- This makes it possible to estimate the occupancy without sampling the signal at the Nyquist rate.
- Thus, D-BigBand tries to detect changes in the occupancy of frequencies that hash to each buckets.
- The authors repeat the bucketization multiple times while randomizing which frequencies hash to which buckets.
- Hence, once the occupancy of a frequency changes, the authors can tell its current state irrespective of its previous state.
6 A USRP-BASED IMPLEMENTATION
- Since the USRPs use the same ADCs, it is not possible to have co-prime sub-sampling rates.
- The USRP digital processing chain cannot support this rate and hence the ADC sampling rate can be set to no higher than 50 MS/s.
- The figure shows the average spectrum occupancy at their geographical location on Friday 01/15/2013 between 1-2pm:, as viewed at a 10 ms granularity (top) and 100µs granularity .
- To collect traces of one GHz of highly occupied spectrum, the authors use many USRPs to transmit and receive.
- The authors run D-BigBand using these sub-sampled versions of the signal.
7.1 Outdoor Spectrum Sensing
- The authors collect outdoor measurements from the roof top of a 24 floor MIT building.
- Fig. 6 shows the fraction of time that each chunk of spectrum between 2 GHz and 2.9 GHz is occupied, as recovered by BigBand.
- These results were confirmed using a spectrum analyzer.
- The figure shows that even frequencies that look 100% occupied over 10 ms windows, become less occupied when viewed over shorter intervals.
- The above implies that the spectrum is sparser at finer time intervals, and provides more opportunities for fine-grained spectrum reuse.
7.2 BigBand vs. Spectrum Scanning
- Most of today's spectrum sensing equipment relies on scanning.
- Here, the authors compare how fast it would take to scan the 900 MHz bandwidth using three techniques: state-ofthe-art spectrum monitors like the RFeye [5] , which is used in the Microsoft spectrum observatory, 3 USRPs sequentially scanning the 900 MHz, or 3 USRPs using BigBand.
- For FFT window sizes lower than 10 ms, the scanning time is about 48 ms.
- Hence, the USRPs spend very little time actually sensing the spectrum, which will lead to a lot of missed signals.
- Of course, state of the art spectrum monitors can do much better.
8.1 Decoding Multiple Transmitters
- The authors verify that BigBand can concurrently decode a large number of transmitters from diverse parts of the spectrum.
- All the transmitters in their implementation use the same technology, but the result naturally generalizes to transmitters using different technologies.
- At any given time instant, each device uses 1 MHz of spectrum to transmit a BPSK signal.
- Note however, that the hopping sequence for different devices allows them to hop to frequencies that get aliased to the same bucket at a particular time instant, and hence collide in BigBand's aliasing filters.
- It shows that BigBand can decode the packets from 30 devices spanning a bandwidth of 900 MHz with a packet loss rate less than 3.5%.
8.2 Signal-to-Noise Ratio
- It is expected that BigBand will have more noise than a narrowband receiver since it can capture a much larger bandwidth.
- This section aims to shed insight on this issue.
- BigBand has higher thermal noise due to bucketization.
- The authors transmit a 10 MHz signal, receive it on BigBand and the narrowband receiver, and compare the resulting SNR.
- At a quantization of 14 bits, the SNR reduction becomes 6 dB which means that the ADC jitter noise is still significantly higher than thermal noise.
9 D-BIGBAND'S SENSING RESULTS
- The authors vary the percentage of total occupied frequencies in the spectrum between 1% to 95% (almost fully occupied).
- As a function of spectrum occupancy, Fig. 10 shows the false positives (i.e., frequencies whose occupancy has not changed, but D-BigBand erroneously declared as changed) and false negatives (i.e., frequencies whose occupancy has changed, but D-BigBand erroneously declares as unchanged).
- The authors see that D-BigBand robustly identifies changes in occupancy, with both the false positive and the false negative probabilities remaining under 0.02 even for a spectrum occupancy of 95%.
10 CONCLUSION
- This paper presents BigBand, a system that enables GHzwide sensing and decoding using commodity radios.
- Empirical evaluation demonstrates that BigBand is able to sense the spectrum stably and dynamically under different sparsity levels; the authors also demonstrate BigBand's effectiveness as a receiver to decode GHz-wide sparse signals.
- The authors believe that BigBand enables multiple applications that would otherwise require expensive and power hungry devices, e.g. realtime spectrum monitoring, dynamic spectrum access, concurrent decoding of multiple transmitters in diverse parts of the spectrum.
Did you find this useful? Give us your feedback
Citations
162 citations
154 citations
110 citations
102 citations
Cites background from "GHz-Wide Sensing and Decoding Using..."
...Strong DVB-T signal reception at channel set S = [22, 23, 25, 26, 28, 29, 30, 33] can be observed in the recorded spectrum....
[...]
...Normalized power spectrum density (PSD) of the real-time TVWS signal recorded at QMUL, S = [22, 23, 25, 26, 28, 29, 30, 33]....
[...]
...needed in compressive wideband spectrum sensing schemes, which do not work well with low-power commodity hardware [20]–[22]....
[...]
78 citations
References
4,812 citations
"GHz-Wide Sensing and Decoding Using..." refers background in this paper
...(Further details of BigBand can be found in our technical report [37] and the complete analysis of the algorithm and proofs of correctness can be found in [13].)...
[...]
1,408 citations
1,247 citations
"GHz-Wide Sensing and Decoding Using..." refers background in this paper
...(Further details of BigBand can be found in our technical report [37] and the complete analysis of the algorithm and proofs of correctness can be found in [13].)...
[...]
1,186 citations
"GHz-Wide Sensing and Decoding Using..." refers methods in this paper
...Once x̂ is computed, it can recover the time signal x and decode the wireless symbols....
[...]
1,138 citations
"GHz-Wide Sensing and Decoding Using..." refers background or methods in this paper
...Some proposals for test equipment reconstruct wideband periodic signals by undersampling [30], [31]....
[...]
...Once x̂ is computed, it can recover the time signal x and decode the wireless symbols....
[...]
Related Papers (5)
Frequently Asked Questions (13)
Q2. What is the key difficulty in capturing GHz of bandwidth in realtime?
The key difficulty in capturing GHz of bandwidth in realtime stems from the need for high-speed analog-to-digital converters (ADCs), which are costly, power hungry, and have a low bit resolution [7], [8].
Q3. What is the basic property of the Fourier transform?
Recall the following basic property of the Fourier transform: sub-sampling in the time domain causes aliasing in the frequency domain.
Q4. What is the way to resolve collisions?
One approach to resolve collisions is to bucketize the spectrum multiple times using aliasing filters with co-prime sampling rates.
Q5. How do the authors use the USRPs to sample the same signal?
In order to sample the same signal using the three USRPs, the authors connect the USRPs to the same antenna using a power splitter but with wires of different lengths in order to introduce small time-shifts.
Q6. What is the motivation behind the FCC’s move to release multiple bands for dynamic sharing?
The rising popularity of wireless communication and the potential of a spectrum shortage have motivated the FCC to take steps towards releasing multiple bands for dynamic spectrum sharing [1].
Q7. What does the paper show about BigBand?
Empirical evaluation demonstrates that BigBand is able to sense the spectrum stably and dynamically under different sparsity levels; the authors also demonstrate BigBand’s effectiveness as a receiver to decode GHz-wide sparse signals.
Q8. What is the definition of an aliasing filter?
an aliasing filter is a form of bucketization in which frequencies equally spaced by an interval BW/p hash to the same bucket, i.e., frequency f hash to bucket i = f mod BW/p, as shown in Fig.
Q9. What would be the cost of decoding all these transmitters without BigBand?
Decoding all these transmitters without BigBand would either require a wideband 0.9 GHz receiver, or a receiver with 30 RF-frontends, both of which would be significantly more costly and powerhungry.
Q10. How does BigBand detect changes in spectrum occupancy?
The authors see that D-BigBand robustly identifies changes in occupancy, with both the false positive and the false negative probabilities remaining under 0.02 even for a spectrum occupancy of 95%.
Q11. How many times do the authors repeat this?
The authors repeat this 4 times at center frequencies that are 250 MHz apart and stitch them together in the frequency domain to capture the full 1 GHz spectrum.
Q12. How many bucketizations are enough to uniquely resolve the colliding frequencies?
Their empirical results (in §7.3) show however that for practical spectrum sparsity (which is about 5%) 3 shifted bucketizations are enough to uniquely resolve the colliding frequencies.
Q13. What is the value of the bucket without a time-shift?
Then the value of the bucket without a time-shift is b̂i = x̂f + x̂f ′ while its value with a time-shift of τ is b̂ (τ) i = x̂f · e 2πj·fτ + x̂f ′ · e 2πj·f ′τ .