1. Support WiFi and LTE Co-existence
Lili Qiu
Department of Computer Science
The University of Texas at Austin

2. Introduction Great success of LTE LTE in unlicensed spectrum
Lack of the bandwidth
Service providers demand more spectrum
LTE in unlicensed spectrum
Proposed by major LTE manufacturers and service providers (e.g., Huawei, Qualcomm, T- Mobile)
Co-existence with WiFi is main issue
Solution – Dividing the spectrum in time-domain
LTE has been in great success over the last few years, and it is expected to keep growing fast. Because of the lack of the bandwidth, the cellular service providers are demanding more bandwidths, and recently, some LTE manufacturers such as Qualcomm and Huawei, and service providers such as T-mobile and NTT are proposing to use even unlicensed spectrum such as 5 GHz channel for LTE. In this case, the main challenge is how to co-exist with WiFi because most of the unlicensed bands are extensively used by WiFi. The main co-existence solution proposed by the manufacturers are dividing the spectrum in time domain so half of the time is allocated for WiFi and another half is allocated for LTE. This cannot be very efficient because we do not know the exact demand of LTE and WiFi traffics.

3. Our Observations Mobile devices have LTE and WiFi antennas
Theoretically feasible to decode concurrent LTE and WiFi signals
Challenges
Decode signals from two heterogeneous PHY
Estimate channel without clean reference signals
Carrier Sense
Synchronization
LTE antenna
In this paper, we propose a different approach for LTE-WiFi co-existence. Our main observation is most of the smartphones have LTE and WiFi antennas, and they are separated enough to have independent channels. Then considering the theory of MIMO, it is possible to decode both of the signals if WiFi and LTE are transmitted together. However, it has many fundamental challenges. First, unlike MIMO that multiple streams are transmitted using the same baseband PHY technique, WiFi and LTE have totally different PHY structure and their signals are not aligned neither in time nor in frequency domain. Second, it is a big challenge how to estimate the channel coefficients without guarantee of the clean reference symbols. Third, there are a few practical issues such as carrier sensing and synchronization.
WiFi antenna
Galaxy S4 antenna location (

4. Proposed idea Let LTE and WiFi send together without coordination
Propose a receiver design that decodes WiFi-LTE overlapped signal
Decode two overlapped OFDM symbols that are not aligned in time or frequency domain
Estimate channel without clean reference symbol
Address practical issues
Carrier sensing
Synchronization
Control-frame reception
This paper mainly focuses on resolving these challenges. We basically allow the LTE base station and WiFi AP transmit together without any coordination, and propose a new receiver design that decodes the WiFi and LTE overlapped signal. The receiver design consists of decoding two overlapped OFDM symbol that are not aligned both in time and frequency domain, channel estimation without clean reference symbol, and solving a few practical issues such as carrier sensing and synchronization. One of our important contributions is we use USRP testbed and perform the experiment in real channel when LTE and WiFi signals are transmitted together.

5. Related Work TIMO [1] – decode WiFi signal under unknown interference
Only decodes WiFi
Clean preamble based WiFi channel estimation
LTE-WiFi coexistence don’t have clean reference signals
ZIMO [2] – decode WiFi and ZigBee signals
Exploit different bandwidth and power in WiFi and ZigBee
LTE and WiFi bandwidths and powers are similar
[1] S. Gollakota, F. Adib, D. Katabi, and S. Seshan. Clearing the RF smog: making robust to cross-technology interference In Proc. of ACM SIGCOMM, 2011.
[2] Y. Yubo, Y. Panlong, L. Xiangyang, T. Yue, Z. Lan, and Y. Lizhao. ZIMO: building cross-technology MIMO to harmonize Zigbee smog with WiFi flash without intervention. In Proc. of ACM MobiCom, 2013.

6. Outline Background Decode when both channels are known
Decode when only channel is known
Decode when neither channel is known
Practical Issues

7. LTE Background Frame transmission is continuous
SF0
SF1
SF4
SF3
SF2
SF5
SF6
SF7
SF9
SF8
10ms
1ms
Reference symbol
Time
Frequency
Frame transmission is continuous
Reference symbols are transmitted periodically
OFDM based transmission (FFT size- 2048)
Actual data bandwidth: 18 MHz
Before describing our idea, let me briefly explain the key characteristics of LTE and WiFi. In LTE, the frame transmission is continuous and there is no idle time between frame transmissions. And the reference symbols that are used for the channel estimation are sent periodically every a few subcarriers and time slots. LTE uses OFDM to address ISI, and when the channel width is set to 20MHz, FFT size is 2048, and the actual data bandwidth excluding the guardband is 18 MHz

8. WiFi Background
Preamble is transmitted only at the beginning of a frame
OFDM based transmission (FFT size - 64)
Actual data bandwidth: MHz
One of the main difference between WiFi and LTE is WiFi transmission is intermittent and distributed and the transmitters accesses channel based on CSMA. Also, reference symbol that is called preamble in WiFi standard is transmitted only in the beginning of the frame transmission. One common thing between LTE and WiFi is they both OFDM, but the FFT size and subcarrier width are very different. Given the channel width is 20MHz, the actual data bandwidth is MHz.
DATA
ACK
Backoff interval

9. Outline Background Decode when both channels are known
Decode when only one channel is known
Decode when neither channel is known
Practical Issues

10. WiFi and LTE overlapped signal
LTE signal
WiFi signal
Received signals
LTE data symbol
𝑠 𝑙 𝑛 = 1 𝑁 𝑘=0 𝑁−1 𝑋 𝑙 [𝑘] 𝑒 𝑗2𝜋𝑘𝑛/𝑁 , N=2048
𝑠 𝑙
𝑠 𝑤
𝑟 1
𝑟 2
𝒉 𝟏 𝒍
𝑠 𝑤 𝑛 = 1 𝑀 𝑘=0 𝑀−1 𝑋 𝑤 [𝑘] 𝑒 𝑗2𝜋𝑘𝑛/𝑀 , M=64
𝒉 𝟐 𝒍
WiFi data symbol
𝒉 𝟏 𝒘
𝒉 𝟐 𝒘
Now let’s talk about how to decode the data when LTE and WiFi signals are overlapped. In LTE and WiFi, because they use OFDM, the data symbols are transmitted in the frequency domain, and in time domain, signals are transmitted after IFFT. At the receiver side, the signals are convoluted with the channel coefficients and LTE and WiFi signals are linearly combined at the two receiver antennas.
𝑟 1 𝑛 = 𝒉 𝟏 𝒍 ∗ 𝑠 𝑙 𝑛 + 𝒉 𝟏 𝒘 ∗ 𝑠 𝑤 𝑛
𝑟 2 𝑛 = 𝒉 𝟐 𝒍 ∗ 𝑠 𝑙 𝑛 + 𝒉 𝟐 𝒘 ∗ 𝑠 𝑤 𝑛

11. Decode when both channels are known
Decode LTE from the overlapped signal
Perform FFT with respect to the FFT size of LTE (2048)
With 2 signals from 2 antennas, decode 𝑋 𝑙 𝑘
Similarly, decode WiFi signal
𝑌 𝑙 𝑘 = 1 𝑁 𝑖=0 𝑁−1 𝑟[𝑖] 𝑒 −𝑗2𝜋𝑖/𝑁
= 𝐻 𝑙 𝑘 𝑋 𝑙 𝑘 + 𝐻 𝑤 𝑘 𝐼 𝑤 [𝑘]
Received signal in the k-th LTE subcarrier
Received signal in time domain
Undecodable WiFi signal
Here I will explain how to decode the LTE and WiFi overlapped signals. let’s first assume all of the channel coefficients are known at the receiver. To decode LTE signal, what the receiver does is performing FFT with respect to the FFT size of the LTE. Then the received symbol at subcarrier k is the channel coefficient of LTE multiplied by LTE data symbol plus the channel coefficient of Wifi multiplied by some unknown WiFi interference. The WiFi data symbols are unorthogonalized because its FFT size is different from LTE FFT size. But what we are interested in here is LTE signal not WiFi signal. With two received signals from 2 antennas, we can get the LTE data symbol by solving two equations with two unknowns. By repeating the same process for WiFi symbol, we can also decode WiFi data symbol. Then the next question is how to estimate the channel coefficients.
𝑌 1 𝑙 𝑘 = 𝐻 1 𝑙 𝑘 𝑋 𝑙 𝑘 + 𝐻 1 𝑤 𝑘 𝐼 𝑤 𝑘
𝑌 2 𝑙 𝑘 = 𝐻 2 𝑙 𝑘 𝑋 𝑙 𝑘 + 𝐻 2 𝑤 𝑘 𝐼 𝑤 𝑘

12. Outline Background Decode when both channels are known
Decode when only one channel is known
Decode when neither channel is known
Practical Issues

13. Decode when only one channel is known
Received WiFi preamble overlapped with LTE signal
Not solvable due to 3 unknowns!
Received LTE reference symbol overlapped with WiFi signal
Obtain WiFi channel ratio from the received LTE reference signal
𝑌 1 𝑤 𝑘 = 𝐻 1 𝑤 𝑘 𝑋 𝑝 𝑤 𝑘 + 𝐻 1 𝑙 𝑘 𝐼 𝑙 𝑘
𝑌 2 𝑤 𝑘 = 𝐻 2 𝑤 𝑘 𝑋 𝑝 𝑤 𝑘 + 𝐻 2 𝑙 𝑘 𝐼 𝑙 𝑘
Known WiFi preamble
𝑌 1 𝑙 𝑘 = 𝐻 1 𝑙 𝑘 𝑋 𝑝 𝑙 𝑘 + 𝐻 1 𝑤 𝑘 𝐼 𝑤 𝑘
𝑌 2 𝑙 𝑘 = 𝐻 2 𝑙 𝑘 𝑋 𝑝 𝑙 𝑘 + 𝐻 2 𝑤 𝑘 𝐼 𝑤 𝑘
𝑌 1 𝑙 𝑘 − 𝐻 1 𝑙 𝑘 𝑋 𝑝 𝑙 𝑘 = 𝐻 1 𝑤 𝑘 𝐼 𝑤 𝑘
𝑌 2 𝑙 𝑘 − 𝐻 2 𝑙 𝑘 𝑋 𝑝 𝑙 𝑘 = 𝐻 2 𝑤 𝑘 𝐼 𝑤 𝑘
Received symbol after removing LTE interference
𝛼 𝑘 = 𝐻 2 𝑤 𝑘 𝐻 1 𝑤 𝑘 = 𝑌 2 𝑘 − 𝐻 2 𝑙 𝑘 𝑋 𝑝 𝑙 𝑘 𝑌 1 𝑘 − 𝐻 2 𝑙 𝑘 𝑋 𝑝 𝑙 𝑘

14. Decode when only one channel is known (cont.)
With the channel ratio, we can decrease the number of unknowns and can solve the equation
Similarly, we estimate LTE channel with known WiFi channel
𝑌 1 𝑤 𝑘 = 𝐻 1 𝑤 𝑘 𝑋 𝑝 𝑤 𝑘 + 𝐻 1 𝑙 𝑘 𝐼 𝑙 𝑘
𝑌 2 𝑤 𝑘 = 𝛼 𝑘 𝐻 1 𝑤 𝑘 𝑋 𝑝 𝑤 𝑘 + 𝐻 2 𝑙 𝑘 𝐼 𝑙 𝑘
𝑌 1 𝑤 𝑘 = 𝐻 1 𝑤 𝑘 𝑋 𝑝 𝑤 𝑘 + 𝐻 1 𝑙 𝑘 𝐼 𝑙 𝑘
𝑌 2 𝑤 𝑘 = 𝐻 2 𝑤 𝑘 𝑋 𝑝 𝑤 𝑘 + 𝐻 2 𝑙 𝑘 𝐼 𝑙 𝑘
𝐻 2 𝑤 𝑘 = 𝛼 𝑘 𝐻 1 𝑤 𝑘
𝐻 1 𝑤 [𝑘] 𝐼 𝑙 𝑘 = 𝑋 𝑝 𝑤 𝑘 𝐻 1 𝑙 𝑘 𝛼 𝑘 𝑋 𝑝 𝑤 𝑘 𝐻 2 𝑙 𝑘 −1 𝑌 1 𝑤 𝑘 𝑌 2 𝑤 𝑘
Given this alpha_k, we can reduce the number of unknowns of the previous equations from 3 to 2, so we can solve the equations and finally we can get the channel of WiFi.

15. Outline Background Decode when both channels are known
Decode when only one channel is known
Decode when neither channel is known
Practical Issues

16. Decode when neither channel is known
How to estimate both LTE and WiFi channels?
Exploit different bandwidth in LTE and WiFi
LTE has 18 MHz and WiFi has MHz  LTE has 1.75 MHz uninterfered channel
Estimate the remaining channel
Extrapolation  inaccurate
Techniques to improve accuracy
Joint WiFi and LTE channel estimation
Iterative channel estimation
In the previous slides, we showed that it’s feasible to get the clean LTE reference symbol and we can estimate LTE channel, and using the estimated LTE channel, we can also estimate WiFi channel. But the main difficulty in getting the LTE channel in advance is it requires us to always decode LTE signal and estimate LTE channel even if we have no data to receive, which can be energy inefficient. Here we introduce the way to jointly decode LTE and WiFi channel that does not require the LTE channel estimation in advance. The main idea is using the channel border that are not overlapped. When the channel width is set to 20MHz, the actual transmission bandwidth of LTE is 18 MHz and WiFi is 16.25MHz. Therefore, at one border of the channel, about 1MHz LTE channels are not overlapped with WiFi channel. Using it, we can estimate the LTE channel of the boundary channel without WiFi interference. However, still most of channels are overlapped with WiFi data. For that part, we use extrapolation and successive channel estimation. Also, to improve the channel estimation accuracy, we iteratively estimate the channel using the data symbol

17. Joint LTE and WiFi channel estimation
Estimate the boundary LTE channel without interference
Estimate the first interfered LTE channel through extrapolation
Estimate the channel of WiFi subcarrier using known LTE channel
Estimate the LTE channels using known WiFi channel
Repeat 2-4 for the remaining part of the half channel
Repeat 1-5 for the other half
LTE Reference signal
WiFi OFDM symbol
Time
Freq.
WiFi subcarrier
LTE subcarrier 1 2 3 4 5
Here I illustrate the joint channel estimation. In the beginning, we first estimate LTE channel in the spectrum boundary without WiFi interference. Second, we estimate the LTE channel of the closest subcarrier through extrapolation. This is feasible because the channel coefficients of the nearby subcarriers has very strong correlation. After that, using the estimated LTE channel, we estimate the channel coefficient for the first subcarrier of the WiFi channel. The method is exactly the same with the estimation of WiFi channel with known LTE channel that I just explained. Once we finish the WiFi channel estimation, using it, we estimate the LTE channel overlapped with the first WiFi subcarrier. Then we repeat the same process to the second WiFi subcarrier and LTE subcarriers overlapped with it, and we repeat it until we estimate the half of the channel. And then we repeat the same job from another side of the channel border.

18. Iterative Decoding
Estimate the channel using WiFi preamble and LTE reference symbol
Decode WiFi data symbols and re-modulate them
Estimate the channel using the re-modulated symbols
The channel estimation using the interfered signals is pretty challenging, so base on our experimental result, that alone does not provide sufficient estimation performance. To improve the accuracy, we use data symbols for the channel estimation. Initially, we estimate the channel using WiFi preamble and LTE reference symbol. Using the initial channel estimation result, we demodulate WiFi symbols and decode them and remodulate them to find the original constellation points of the data symbols. After that, using the WiFi data symbols, we perform the channel estimation again. This is helpful improving the accuracy because WiFi has only 2 known OFDM symbols that can be used for the channel estimation. But if we use data symbols for the estimation, we can get a few hundreds of known symbols in addition for the channel estimation. But it can be a tradeoff because it increases the complexity. In the performance evaluation, we show how many data symbols are required to get the sufficient channel estimation accuracy.

19. Practical considerations
Decode WiFi MIMO signals
Our design is extendable to WiFi MIMO with multiple WiFi antennas and additional LTE antenna
Carrier sense only WiFi signals
Project the received signal onto the null space of the LTE channel
Synchronization for LTE
Use WiFi null DC subcarrier and LTE synchronization sequence (PSS, SSS)
Different channel widths
Nullify a few data subcarriers for channel estimation purpose
In this talk, the main issue is decoding the interfered signal and the channel estimation. Besides it, there are several practical issues to enable LTE and WiFi coexistence in unlicensed band. I’m not going to introduce them in detail today, but the practical issues that we address in the paper are decoding wifi MIMO signals, carrier sensing, synchronization, and how to handle when LTE and WiFi channel width is not the same. You can find the detail in the paper.

20. Performance Evaluation
USRP testbed
10 MHz channel
USRP nodes at 8 locations with different channel conditions
Use the best rate at each location
Now I will introduce the performance evaluation result. For the experiment in real channel, we set up the testbed with USRP software radio. Because of the limitation of the processing power, we set the channel bandwidth to 10MHz. We evaluated the throughput performance and the accuracy of the channel estimation.
the best modulation is used among BPSK, QPSK, 16QAM for WiFi, and QPSK and 16QAM for LTE

21. Compare with TDMA 1.87x throughput gain from time-division
The first result is the comparison with time division based approach. In time division approach, half of the time is used for LTE and the other half is used for WiFi. In terms of the average throughput across all locations, our approach gets 87% higher throughput compared to the time division approach. It is slightly less than 2 because of the imperfect channel estimation.
1.87x throughput gain from time-division

22. Compare different channel estimations
Here I compare the throughput with different channel estimation method. Compared to the channel estimation with clean LTE reference WiFi preamble, our scheme has slight throughput reduction. The throughput of our scheme is 93% of the throughput of known LTE and Wifi preamble. Here ‘extrapolation only’ is estimating the channel only using the boundary channel and only relies on the extrapolation to estimate the other interfered channels. Because the channel condition we performed the experiment was frequency-selective, the performance of the extrapolation is not satisfactory.
96% throughput compared to known LTE channel
93% throughput compared to known LTE and WiFi channel

23. Estimate channel using different # WiFi symbols
Here we evaluate how many symbols are required for the channel estimation with sufficient accuracy. This graph shows the throughput in 8 locations with different number of data symbols for decoding. If you the result, you can see throughput is a little low when 20 symbols are used for the channel estimation. If more than 50 symbols are used, the throughput is not distinguishable. Therefore, we conclude 50 symbols are sufficient for the accurate channel estimation.
50 data symbols are sufficient for channel estimation

24. Conclusion
A novel coexistence mechanism for LTE and WiFi in unlicensed band
Decoding data symbol under cross-technology interference
Channel estimation without clean reference
Address practical issues
In real channel experiment with USRP
90% throughput improvement compared to time-division
7% throughput loss compared to clean reference based decoding

25. Thank you!

26. Decode WiFi MIMO
In WiFi MIMO, one cannot get WiFi channel ratio due to LTE and WiFi interference
We let only one of WiFi antennas send the signal during LTE reference symbol transmission
In n/ac, each TX antenna separately transmits preamble
Using the interfered preamble and the channel ratio, we can apply the same method to estimate WiFi channel

27. Carrier Sense Only WiFi Signals
Due to continuous LTE transmissions, we should only carrier sense WiFi signals
Projection based WiFi carrier sensing
Estimate LTE channel using LTE reference symbols
Project the received signal onto the null space of the LTE channel
If only LTE signals exist, the projected signal is small
Otherwise, there is WiFi transmission and we need to defer
What if LTE channel estimation is wrong due to WiFi interference?  WiFi transmission detected!

28. LTE synchronization under WiFi interference
LTE transmits Primary Synchronization Sequence (PSS) and Secondary Synchronization Sequence (SSS) every 5 ms for synchronization
960 KHz bandwidth
One third of PSS overlaps with WiFi DC subcarrier where no signal is transmitted
With 30% of non-overlapped bandwidth, we can detect LTE signal timing even under -7dB SINR

29. Wider channels
If WiFi chanenl > 20MHz, we cannot rely on guardband difference to get clean LTE reference symbols
Solution
Nullify a few WiFi subcarriers and use them for LTE channel estimation
Nullify 3 subcarriers every 10 MHz is sufficient
10% bandwidth loss for WiFi
in return allows concurrent LTE transmissions