HKUST January 3, 2007 Merging Propagation Physics, Theory and Hardware in Wireless Ada Poon University of Illinois at Urbana-Champaign
Outline Multiple-antenna (MIMO) channels Human body wireless channels
Toy Problem Putting a lot of antennas close together yields arbitrary large gain?
Conventional spatial degrees of freedom does not take into account space For an multiple-antenna system, the number of spatial degrees of freedom is under suitable channel fading conditions.
Is the number of antennas fundamental? Dimension Frequency Space Degrees of freedom Theory Implementation 2W N f S N N antennas ADC ADC ADC f S
Spatial Channel Models Multi-input multi-output (MIMO) model Models the response of all antenna pairs Proposed model Models the response of antenna arrays with physical area limitation Models the angular location and extent of physical paths Ray-tracing model Models the response of all physical paths More physical description Analytical tractability
Measurement Environments Office and townhouse Intel office Stationary Automated positioner STA AP
Angular-domain is more interesting! 8 8 2e-3 Frequency (GHz) 7 6 5 4 Frequency (GHz) 7 6 5 4 1.5e-3 1e-3 3 3 2 0 5 10 15 20 25 Antenna Position (cm) 2 0 36 72 108 144 180 θ 0.5e-3
Clustered phenomenon is observed! Channel response 0.1 Back-wall reflection Scattering from furniture 0 0.1 0.2 20 15 0 50 100 150 200 250 300 350 10 5 Time of arrival (ns) Angle of arrival ( ) It occurs over a wide range of frequencies and is consistent with measurements in the literature. Typically, there are 1 to 5 clusters. The angular extent of each cluster ranges from 20 to 30 in indoor channels.
Array Response with Area Constraint p Surface Electric dyadic Green function in free space In the far field, the response is Area constraint is measured by
Clustered Channel Response Channel scattering is captured by Scattering condition is measured by the total solid angle TX/RX 1 m In a fully scattered environment, is a sphere and
Main Result I Transmit and receive signal spaces: Number of spatial degrees of freedom is Polarization gain Effective area of antenna array Total solid angle of scattering clusters A. Poon, R. Brodersen, and D. Tse, Degrees of freedom in multiple-antenna channels: a signal space approach, IEEE Trans. Inform. Theory, vol. 51, pp. 523 536, Feb. 2005.
Packing Beams over Channel Solid Angles φ θ Spatial degrees of freedom
Where is the factor of 2? Total number of spatial degrees of freedom is 4. A factor of 2 is from resolving the propagation direction. Another factor of 2 is from resolving the oscillation direction polarization benefit. Scattering will not increase spatial degrees of freedom from polarization.
Choosing the Number of Antennas unit sphere Continuous linear array Discrete linear array of length and width No. of antennas ( antenna spacing) No. of antennas No. of antennas 0 0 no information loss 0 overlapped information loss 0
Choosing # of antennas on Planar Arrays no information loss Scattering condition perceived by a continuous planar array No. of antennas area of hexagon area of circle No. of antennas
A System View For example, typical office environment has 3 clusters, each of angle 30, On knowing of the application environment helps optimize number of antennas plus RF/analog front-ends. It is analogous to knowing the delay spread to optimize Number of taps in the channel equalizer of DS-CDMA systems Number of sub-carriers in the cyclic prefix of OFDM systems.
Capacity Analysis Need to model the channel response in more detail over Consider linear arrays and the elevation direction. Two scattering scenarios : Specular reflection Diffused scattering
Main Result II At high SNR, where L is the array length and M is the number of scattering clusters. A. Poon, D. Tse, and R. Brodersen, Impact of scattering on the capacity, diversity, and propagation range of multiple-antenna channels, IEEE Trans. Inform. Theory, Mar. 2006.
Spatial degrees of freedom depends on SNR Dependence is logarithmic in SNR If we are very desperate on the degrees of freedom, we could pack more antennas and operate at a higher SNR. Physical interpretation from superdirectivity:
Angular-domain Processing Very similar to classic array processing. In array processing, we would estimate Number of sources Directions of arrival of sources Now, we estimate Total angle spread Angular intervals subtended by scatterers However, array processing techniques usually assume uniform antenna spacing, regular array geometry, and unambiguous array manifold. Use multiple-window idea to derive an angular spectrum estimator that is robust against any array configuration.
Total DOFs over Time/Freq/Space Capacity (bits per frame) Duration Bandwidth Scattering condition Array area Polarization
Perspective on Future Flexible Radios Traditional wisdom on wireless builds on the assumption that there are limited degrees of freedom (over time, frequency, and space) simultaneously shared by many users. Develops multi-access schemes that can squeeze as many bits as possible over the limited resources. In the open spectrum environment, there are plenty of degrees of freedom shared by a few users. Develops agile communication systems that can sense and learn the environment, then find and adapt to the sweet spot.
Agile Communication Systems MIMO antenna systems to improve the sensitive of RF frontend and to avoid interference Mixed-signal MIMO receivers in CMOS More aggressive partition of computation fabrics (programmable cores, special-purpose engines, and dedicated logic) to avoid providing excessive flexibility in baseband processing Reconfigurable baseband processor
Reconfigurable Architectures Most existing reconfigurable architectures focus on the computation model. But energy efficiency is mostly driven by the granularity of configurable units. We focus on increasing the granularity without compromise flexibility by matching the granularity of configurable units to the degree-offreedom (DOF) processing in most wireless systems.
Processor Macro-architecture Mem ory Input inter connect DOF DOF DOF DOF DOF DOF DOF DOF ML DOF DOF DOF DOF ALU Output inter connect Configurable Units DOF (dominant) CORDIC ML Dual-core ALU Operations Supported FIR filter Auto-correlation Cross-correlation De-spreading N-point FFT Matrix-vector multiply Euclidean distance calculation sine, cosine asine, acosine Normalization Maximum likelihood Miscellaneous math operations Asynchronous control
Performance of Prototype Processor We have programmed the processor to run Radix-4 FFT Two streams of simultaneous radix-2 FFT (for MIMO systems) FIR filters Time and carrier synchronization algorithms Intel 0.13 um CMOS, 500k gates Clock frequency = 200 MHz Power = 63.4 mw Energy efficiency = 95 MOP/mW Energy efficiency is on the same order as fully dedicated architecture. A. Poon, An energy-efficient reconfigurable baseband processor for wireless communications, submitted to IEEE Trans. VLSI Systems, revised Sept. 2006.
Popular MIMO Receiver Architecture LPF A/D LNA 90 Cx Mult LPF A/D Cx Add LPF A/D LNA 90 LPF A/D Cx Mult DSP Large number of analog components Strong interferers easily saturate the ADCs, rendering any baseband MIMO algorithm useless. However, most of current research activities focus on the VLSI implementation of baseband MIMO algorithms.
RF/Analog-based Architecture VGA LNA VGA 90 LPF A/D DSP VGA LPF A/D LNA VGA 90 to VGAs Combine antenna outputs in the RF/analog domain to reduce the number of components and to enhance sensitivity to the intended user. Requires RF VGAs which are commonly implemented in non-cmos process and consume significant amount of power. Not suitable for fully-integrated and low-power applications
Proposed Mixed-Signal Architecture LNA Complex Multiplier LPF A/D DSP LNA Complex Multiplier LPF A/D Quantization Algorithm An algorithmic approach The complex multiplier is a bank of mixers with multi-phase inputs. Outputs from the quantizer select the way to combine the mixer outputs. The quantization algorithm is based on frame theory and stochastic approximation. We are planning to tapeout a test chip of the complex multiplier at 90 nm CMOS in summer.
Human Body Wireless Channel [Meng, Neurons to Silicon: Neural Interface Design, ISSCC 2006.]
Conclusions Incorporation of physical concepts to understand multiple-antenna channels is necessary for the design of more efficient wireless systems. Processing in the angular domain reduces computation complexity, feedback rate, and possibly leverage cross-layer schemes. Matching the granularity of configurable units to DOF processing yields an energy-efficient and flexible architecture. Applying algorithmic ideas to leverage RF/analog implementation of multiple-antenna systems. Bio-networks present new challenges that do not exist in free-space wireless communications. A thorough understanding requires a profound grasp of the interplay of communication theory, circuits, and electromagnetics.