Improving Wireless Access Technologies Adam Wolisz Professor of EE&CS, Technische Universität Berlin, TKN Adjunct Professor, EE&CS Dept, UC Berkeley, BWRC http://www.tkn.tu-berlin.de/~wolisz/wolisz.html Sept.26,2007

Overview Short introduction of the TU Berlin research environment Towards Dynamic OFDM (OFDMA) Dynamic spectrum usage with Cognitive Radios

Acknowledgements: The contribution of my collaborators/students notably TUBerlin: Dr. Gross, Mathias Bohge, Oscar Punal, Daniel Willkomm, Murad Abusbeih UCBerkeley: Prof. Brodersen, Dr. Cabric, Mubaraq Mishra ST Microelectronics: Wendong HU, Dr. George Vlantis Is gratefully acknowledged.

Main-Campus Berlin University of Technology

TU Berlin: The Faculty of EE&CS Established in April 2001 as a pilot fusion of EE and CS. 43 associate or full professors (German C4 or C3) + a numerous “Professors in Residence” 3 curricula (Number of beginners/Year, trend) EE (260+), CE (150+), CS (300+-) Communication Technology is one of the major focus areas – for the whole TUB.

Communications Research relevant environment The Institutes in 2004 total budget of € 165m, (€110m in external grants) 1800 employees, incl. 380 employees with a PhD 2 1 2 1 3 Fraunhofer-Gesellschaft COOPERATION: 21 joint faculty appointments; incl.17 fully funded by partners 2 TU Berlin 4 2 1 1 1 1

Telecommunication Networks Group (TKN) People: Full professor (Chair) A. Wolisz + (2-4) Post-Docs / Assistant Professors + (20-25) Research Assistants/PhD students + 3 Technical Staff + 2 administrative assistants Supported by grants from: EU, BMBF, DFG, DAAD, Siemens AG, Ericsson, DoCoMo, .... External Funding – approx. 1.5 Mio €/Year Som-2006e numbers for the time 2000 9 post-doc research associates/Phd Graduates/adjunct lectureres have been appointed professors. Degrees granted: Doctorate: 20 , Diplomae: Over 90 Papers: ca. 50 in journals, magazines or book chapters ca. 150 in refereed conferences/workshops see www.tkn.tu-berlin.de/publications

Research Topics (selection) General Direction: Architectures and Protocols for Networks Optical backbone/ Optical Metro Networks... Wireless access...(QoS + Capacity). Mobility incl. Group mobility and High speed mobility Sensor networks Cognitive Radio

The Fundamental Problem Wireless dominates last hop (s?) Because cable is always a constraint... Two fundamental features of wireless communication: Interference; i.e. influence of any transmission on other ones calling for proper Separation (space, Frequency, time, code) Dynamic change of the received signal strength in spite of constant transmission power even without interference. These features result in challenges: Limited Capacity, given frequency spectrum and space. Difficulties in proper QoS support even on a single link

Introduction: OFDM Orthogonal Frequency Division Multiplexing (OFDM) splits the bandwidth into narrowband sub-carriers Parallel symbol transmission (reduces intersymbol interference) Orthogonal sub-carriers (no intercarrier interference) Fading produces a strong variation in the sub-carrier gains: always some sub-carriers in a “bad state” Orthogonal Frequency Division Multiplexing This changes in the time domain as well…

OFDM as transmission scheme! Basic Scenario Access Point Backbone Terminals Downlink Data Queues at Access Point OFDM as transmission scheme!

Link Adaptation IEEE 802.11 a/g : Time division Multiple Access and adaptive modulation/coding - the same over all sub-carriers. Average channel gain adaptation: the few sub-carriers with the lowest gain dominate the BER & PER [Awoniyi06] Same Modulation and Coding Scheme on all Sub-carriers, which are assigned to one station

Adaptive Modulation Scheme For each individual sub-carrier the selected modulation schema assures the highest bit-rate for upper-bounded BER.

Dynamic OFDM: Adaptive Modulation Adapts the modulation type to the current gain of each sub-carrier subject to a bit error probability target, performing adaptation on a per-packet base. Theoretically this procedure has been shown to outperform Link Adaptation [Czylwik98] BPSK Sub-Carrier 1 16-QAM Sub-Carrier 2 Sub-Carrier 3 Sub-Carrier 4 64-QAM Sub-Carrier 43 QPSK Sub-Carrier 44 Sub-Carrier 45 Sub-Carrier 46 Sub-Carrier 47 Sub-Carrier 48 STA 1 Assignments: NO MOD.  SNRx ≤ Xi BPSK  Xi < SNRx < Xj QPSK  Xj ≤ SNRx < Xk …

Dynamic OFDMA Performance Improvement [Wong99] The whole set of sub-carriers is split into sub-groups, which are then assigned to different stations in a FDM fashion. STA 2 STA 1 STA 3 BPSK Sub-Carrier 1 16-QAM Sub-Carrier 2 Sub-Carrier 3 Sub-Carrier 4 64-QAM Sub-Carrier 43 QPSK Sub-Carrier 44 Sub-Carrier 45 Sub-Carrier 46 Sub-Carrier 47 Sub-Carrier 48 MULTI-USER DIVERSITY

Now: Optimization approach Open question: How to generate the subsets of sub-carriers serving individual flows? Two-step approach (Yin et al. 2000) per time SLOT. First: sub-carrier allocation Determine the number of sub-carriers for each subset (meeting the demand) IDEA: Utilize packet queue… Second: sub-carrier assignment Choose sub-carriers according to the allocated number for each subset IDEA: Utilize channel-related information for the assignment (goal: increase the capacity, assure fairness)

Allocation: satisfying the demand... Let us assume ONE Queue per terminal (i.e. flow) The queue will built up if bad transmission ... Packets should be dropped if waiting excessively. Observation: Not all frames in an MPEG video are equally important (I, B, P frames) Drop packets based on importance: I frames on deadline, P frames 25% earlier, B frames 50% earlier. Allocate the bit-rates (simplified: NUMBER of sub-carriers) based on weighted lengths of queues The size of important packets is given a larger weight ... Queue length is the sum of these weights

Assignment: Integer Programming Formulation More flexibility by formulating assignment as Integer Program {1,...,J} : Set of wireless terminals, {1,...,S} : Set of sub-carriers gj,s: CNR of terminal j on sub-carrier s ps : power assignment for sub-carrier s F(): Mapping of subcarrier SNR to applied modulation type cj,s:(= 0,1): Assignment of sub-carrier s to terminal j zj: Subcarrier allocation for terminal j Basic formulation: Each subcarrier only assigned once Maximize capacity Number of sub-carriers per terminal fixed out of Allocation!

Optimal solution (constant power assignments) Assignment problem maps to a graph theoretical problem Maximum weight bipartite matching problem An optimal algorithm for this problem exists – the Hungarian algorithm with complexity of O(S3) Measured run times of the algorithm: ~2ms are too long compared to the Allocation/Assignment TIME SLOTS < coherence time of wireless channels in our test data Good heuristics exist for this problem, solving the problem with average parameter setting within 500µs Performance loss:  10%

A heurisitc for the Assignment... Sub-Carrier Weight: 1 3 3 2 2 3 3 3 2 1 2 1 Sub-Carriers Single Sub-Carrier States towards: 1 2 3 4 5 6 7 8 9 10 11 12 User 1 G G G G B G G G G G B B User 2 B G G G G G G G G B G G User 3 B G G B G G G G B B G B Assigned Sub-Carrier Set of User 1 Assigned Sub-Carrier Set of User 2 Assigned Sub-Carrier Set of User 3 Dynamic FDM assignment of sets of sub-carriers to individual terminals

Numerous results addressing: Joint optimization of power (loading) and assignment. Including the overhead for signaling (how should the receiver now which sub-carriers have been assigned?) Allocation including priorities See our URL… =================================================================== But: does this require completely new systems? We suggest the Adaptive Modulation in the up-link and Dynamic OFDMA in the down-link… A proposal for including this to IEEE 802.11… as backward compatible solution. - Technical Report: TKN-07-002 - Submissions to IEEE 802.11, VHT SG

Required changes to use Dynamic OFDM In order to choose an (optimal) modulation/coding per sub-carrier, we need to - - estimate the channel gain per sub-carrier for each transmission Obligatory RTS/CTS - inform the receiver about modulation/coding used per sub-carrier  extended header in data packets - adjust the NAV settings after the transmission For the multi-user case (parallel transmission of packets), we have to add Multiple CTS per RTS and ACKS per data packet - multtiple CTS per RTS, and ACKS per data packet; signal the assignment of sub-carrier sets, Higher Performance  but also higher processing power.

Schemata of the changes Single user Multi-user

Some performance data (simulation!) • Simulation Settings: - Saturation mode, - 4 and 8 stations, - downlink traffic • Scenarios: – Multi-user mode – Single-user mode with Round Robin Scheduling – 802.11 a/g with RTS/CTS and Round Robin Scheduling • Metrics: – Goodput, PER and PHY Efficiency

This approach has potential… Goodput - 8 STAs - Large Packets (1564 Byte) - 802.11 a/g with RTS/CTS Goodput - 4 STAs - Large Packets (1564 Byte) - 802.11 a/g with RTS/CTS

Ongoing work … We have proposed recently a proposal for the interface definition and API between MAC-PHY 3FPP LTE simulator (Ericsson) – used for sensitivity of the system efficiency on the Control Channel error due to interference from neighbor cells… The real overhead, inaccuracies of sub-channel estimations, etc. require experimental investigation. Several PHY developments (FPGA based) in preparation. Execution of the algorithms on FPGA platforms is to be considered.

The underlying philosophy: Using the best opportunity. Why not use the same philosophy for grabbing more spectrum… This is Cognitive Radio based spectrum utilization…

Nutrition facts: The bad news... The amount of non-licensed spectrum (in interesting frequencies) is limited  3 4 5 6 GHz Allocation in the 3GHz-6GHz

Nutrition facts: The good news... Licensed users do not REALLY use ALL THEIR spectrum ALL the time in EACH place the license holds  The estimates are like 80%? 90%? 99%? unused... A snapshot from Berkeley: real measurements…

Re-Usage of sectrum: What are the options? The licensed users – called PRIMARY USERS – could be screened, and Be permitted to “sell” or “lease” their licenses…. Loose licenses (for some area) if spectrum not used properly Have licenses limited to some times of the day “recycled” assignments could be opened as ISM bands The primary user (or a “broker” on their behalf) could be obliged to “announce” unused frequencies (again in space and time domain); usage as ISM bands is possible. A new category of users - SECONDARY USERS - possess the ability to assess autonomously the temporarily unused spectrum and grab it for “specific” usage without primary users being aware of this “kidnapping”

A Primary User X … legally owns some frequency band … can tolerate a maximal interference time tx each time he resumes channel usage A secondary user system has tx time units to detect a primary user and clear the corresponding Sub-Channels (time domain!) tx is dependent on the primary user system and may vary from system to system … is not (cognitive) secondary radio aware (i.e. does not provide specially signaling of activity – especially no preparation to re-gain his frequency band) NOTE: IF primary user would use a carrier sensing protocol, neglecting the (unknown!) secondary user MUST be assured (operation below the carrier sense sensitivity!)

Cognitive radio for “Secondary Users”

Sensing while sending??? (a system view) A fundamental challenge … three possible answers Interrupted sending… (see e.g. IEEE 802.22 basic…) Quick sensing needed. Bad for QoS of the secondary's Part of the “primary” band not re-used (see Corvus) Coordination of frequency usage required Band only partially available for sensing…

Interleaved frequency usage (see CORVUS) Primary user frequency band (F-Band) Active primary user Hz Sub-channel Secondary user link (SUL) Bandwidth B [Hz] Divided into N sub-channels of bandwidth b=B/N [Hz] PU F-Band covers multiple sub-channels Sub-channels of active Primary users (Pus) can't be used by SUs SUs compose SU-Links out of free sub-channels (Re-)appearance of PU All affected Sub-Channels have to be cleared New sub-channels should be acquired

This resembles OFDMA… sure: Efficiency: Potential for usage of the spatial diversity (anyway in downlink) Available actual bandwidth of each sub-carrier (depending on the channel conditions) fully utilized. Constant Sub-carrier Spacing Robust to channel positioning (offset) and bandwidth changes Interesting options for usage for the secondary devices in the “Interleaved transmission” modus.

Sensing while sending??? (a system view) cont. A fundamental challenge … three possible answers Interrupted sending… (see e.g. IEEE 802.22 basic…) Quick sensing needed. Bad for QoS of the secondaries Part of the “primary” band not re-used (see Corvus) Coordination of fre-quency usage required. Band only partially available for sensing… Regular hopping (see e.g. IEEE 802.22 DFH mode) Changing frequency bands regular event (collision in 802.3) “Relaxed” sensing in free(?) band – assuming coordination

IEEE 802.22 (DFH) Worldwide first draft of a Cognitive Radio standard Provide wireless broadband Internet access using TV-bands Ensure non interference with incumbents (grace period 2s) through spectrum sensing - Philosophy following WiMax (IEEE 802.16) Basic Mode: Assure respecting the grace period by sensing during TRANSMISSION INTERRUPTION DFH: Perform data transmission and sensing in parallel Transmit data on channel X and perform sensing on channel Y After 2 seconds channel Y is used for data transmission and the next channel is sensed. Might be X again...2 channels/cell Thus an 802.22 cell hops through a set of working channels

DFHC (DFH Communities) – some ideas Each community has a community leader Leader is selected through leader election Leader calculates a hopping pattern for each member of the community (i.e. cell) Leader is responsible for accepting / rejecting new members Neighborhood discovery One-hop broadcast of used frequencies and current interference situation by all 802.22 cells Used to create and maintain communities It is possible to support N Cells with (N+1) frequencies…

DFHC hopping pattern W. Hu, D. Willkomm, L. Chu, M. Abusubaih, J. Gross, G. Vlantis, M. Gerla, and A. Wolisz, "Dynamic Frequency Hopping Communities for Efficient IEEE 802.22 Operation", IEEE Communications Magazine, Special Issue: "Cognitive Radios for Dynamic Spectrum Access", May 2007 Cells need to shift their operation periods by one quiet time Quiet time is the minimum time needed to sense a channel WRAN2 WRAN1 WRAN3

Protocol sketch for schedule maintenance Hopping patterns can change Due to incumbent appearance on a used channel Due to a member leaving / joining a community Community leader needs to calculate new hopping pattern and distribute it in the community Consistency issue: How to assure that all members receive the new hopping information If not all members switch to the new hopping pattern simultaneously there might be collisions between the old and the new hopping pattern Solution: Hopping pattern lifetime and sequential switching

Hopping pattern lifetime Hopping patterns have a specific lifetime After expiration of the lifetime the hopping pattern cannot be used anymore The leader thus has to periodically renew the hopping pattern Upon renewal the pattern can be changed, new members can be added, etc. This ensures consistency: even if some members do not receive the new pattern, they cannot use the old one anymore But what if a hopping pattern needs to be changed in the middle of a lifetime (i.e. due to appearance of an incumbent)? Solution: Sequential Switching

Sequential Switching The leader sequentially switches all members one by one to the new hopping pattern New hopping pattern is collision free with the pattern of all members not switched yet Implicit acknowledgement: sensing on the newly assigned channel (implicit confirmation by acting) Even if somebody fails to follow - all members already switched can use the new pattern without any collisions

In any case: Detection of (possibly) Weak Signals Cognitive Radio users must guarantee non-interference requirement Primary User Tx CR(Tx) Rx CR(RX) Decoding SNR Sensing SNR distance Distance and channel not known Cognitive radio observes (senses) primary system signals Those might be strongly attenuated While the transmission of the CR(Tx) towards Rx is not… In the order of … This is a noise dominated regime, that we did not address so far in communication systems

Solution: Network Spectrum Sensing [BWRC,Cabric] If spacing >> λ/2 a few cooperative radios give big improvements Prob. of detection 5 radios 1 radio Prob. of false alarm

We need a signaling channel: Cooperation is needed… at least in “proximity” For assuring that “no other secondary is transmitting during the sensing” For assuring network spectrum sensing. Who should be subject to coordination? How to organize the exchange of information for coordination??

But: Control Channel needed (logically) for: Universal Control Channel (UCC) Globally unique Used to get necessary information for creation of new groups and to announce them Used by new users to choose and join a specific Group Group Control Channel (GCC) Each SUG has own control channel Used for exchange of sensing information – recognition of primary users. Used for data channel establishment (out of temporarily available resources) and its maintenance in spite of re-appearing primary users. Numerous claims in favor of a specific Control Channel are recently being made…

Options for the control channel implementation An Universally/regionally pre-assigned frequency…. Globally unique Will be difficult…. ISM band… Globally available… What about possible (strong??) interference? UWB Not interfering Tradeoff: distance vs. bit-rate might be very useful Cost? Deployment? One of the “available channels” Convention for selection needed IEEE 802.22 considers this variant….

BWRC Platform [BWRC: Cabric, Tkaczenko] Sensing PHY real-time processor: 4 FPGAs ~ 10M gates ASIC at 250 MHz On-chip memory: Soft+Hard > 10 Mbits Dynamic Partial Reconfiguration Dedicated DSP blocks: 18b mult + MAC Architecture optimization for ASIC - Parallelism/Pipelining/Interleaving - Bitwidth optimization - Area estimate: 10,000 slice = 1mm2 Handle multiple radios on one board. Sensing MAC embedded processor: Central FPGA: Linux + Full IP Embedded processors: PPC+ARM On-chip Ethernet MAC Bus connection to 4 other FPGAs Radio interfaces: 16 high speed radio links (10 Gbps) 4 interfaces per FPGA Fiber optic cable compatible

Reconfigurable Wireless Radio Modem [BWRC] Suitable for sensing and transmission in TDD mode A/D 12b/64MHz D/A 14b/128MHz 2.4 GHz radio (85 MHz) ISM band Sensing radio processor 10 Gbps infiniband connection supports fiber optic cables Antenna

Why 2.4GHz? Very crowded spectrum with unlicensed devices. IEEE 802.11 b/g cards within laptops, are quite programmable and allow users to control their transmission parameters. Easy to implement protocols on these cards Hardware and software support for the 2.4GHz bands is already developed within BWRC BWRC cards can be programmed to the complete 80MHz band

First Experimental setup at 2.4GHz [BWRC/TKN] BEE2 radios perform sensing Laptops perform transmission Laptops connect to the BEE2 via standard TCP/IP

What about control channel ?? In first set-up: “Internal communication” between sensing boards Ethernet communication between the laptops… One selected wireless channel, ot IEEE 802.11.a could also have been used. This might be enough for SOME of the experiments… ================================== Let us also keep in mind some (available) other option, the IEEE 802.15.4a CHIRP solution (Nanotron, Berlin) Promises: A pretty robust transmission with range sufficient for WLAN type deployment…. + inherent (precise) LOCATION Samples being in possession of TUBerlin/TKN for evaluation….

Functions of „Classical Wireless Systems“ We will consider here: Cellular, infrastructure WLANs, ad-hoc networks… What has been typical for all of them? (history) Usage of specific frequency bands  optimized transmission Exclusive usage of frequency bands  by regulation Channel structure within this bands - using of a selected /negotiated channels for the whole transmission  assigned at the beginning of the data exchange Basic steps Finding the partner  mostly by beaconing (notably base stations!) Exchanging set-up data  separate signaling channel or in-band Selecting a channel to work on  ditto

What does change in CR ? (potentially) Usage of arbitrary frequency ranges (for much better spectrum utilization) Approach: Consulting data basis + distributed sensing No fixed channel structure (for flexible adaptation to the traffic needs…) Approach: Negotiating variable channel structure Imposed high dynamics in changing the used frequencies (if the primary user pops-up!) Approach: sense while communicating (how?) Usage of arbitrary transmission schemata Approach: Use OFDMA  (see 802.22)

Looks like a revolution? Not entirely… Usage of specific frequency bands ? GSM: 900/1800/1900 …. IEEE 802.11: 2.4GHz, 5.4 GHz, 5.9 GHZ (DSRC)… WiMAX… Specialized transmission schemata? General trend towards OFDM /OFDMA!!! Exclusive usage of frequency bands? Take 802.11 …. Coexistence with radars (in 5 GHZ/Europe), microwave ovens, Bluetooth, 802.15.4, etc Persistent usage of assigned channel? Some frequency hopping during the communication (802.11 FH, Bluetooth, Some GSM) Channel change in 802.11 (e.g. Mobility…)

Lessons learned: There is a trend to increase the dynamics of adaptive resource usage on each time scale and granularity. OFDMA seems to be especially attractive transmission schema Available systems/set-ups allow for investigation of multiple features for the future “Cognitive Radio”

Thank You !

IEEE 802.15.4a Chirp … some facts [www.nanotron.com] Chirp technology operating in 2.4 GHz ISM band, 20 MHz, 7 channels (3 non overlapping) Data rates 2, 1 Mbps; 500, 250, 125, 62.5, 31.25 kbps Power/sensitivity Tx: up o dBm (plus ext. amplifier for long range) Rx: -97dBm @ 250kbps; BER= 10-3 (with FEC) Symmetrical double-sided two way ranging 128 bit hardware encryption. RS 232 interface (USB expected…)

IEEE 802.15.4a Chirp … early data [IEEE Documents] Indoor (European office building) measurement results for a data rate of 1 Mbps (over air interface) and BER = 0.001 are as follows: Output power (EIRP) = -30 dBm (1 µW), distance = 5 m, 1 wall Output power (EIRP) = -15 dBm (32 µW), distance = 23 m, 4 walls Outdoor measurement results for a data rate of 1 Mbps (over air interface) and BER = 0.001 are as follows: Output power (EIRP) = +7 dBm (5 mW), distance = 739 m (+/-10 m) Both transceivers use equivalent isotropic antenna (gain = 0 dBi) For long ranges the transmit power may be allowed to rise to each country’s regulatory limit For example the US would allow 30 dBm of output power with up to a 6 dB gain antenna The European ETS limits would specify 20 dBm of output power with a 0 dB gain antenna Ranging: 2m indoors – 1 m outdoors…. (data sheet)

Communication Stack PHY Layer Link UCC GCC Data Transfer Channel Secondary users create communicating GROUPS (SUGs) A universal control channel as well as group control channels are assumed... PHY Layer Link UCC GCC Data Transfer Channel Spectrum Sensing Channel Estimation Data Transmission MAC Group Management Link Management Data Transmission Control channels....