Signal processing is an art that deals with the representation, transformation and manipulation of the signals and the information they contain based on their specific characteristics. The second part of the thesis focuses on signal processing algorithms for data compression and filter bank designs.
MIMO Transceiver Optimization
MIMO Channel Models
Two of the most common techniques for this are the zero-padding and cyclic prefix precoding techniques [89]. The zero-padding precoding produces the effective frequency flat channel matrix H with Toeplitz structure, and the cyclic-prefix precoding produces circular matrix H.
Transceiver Optimization and History
Many of the existing works were shown as special cases within this framework. Most of the results described above assume that the channel, precoder, and equalizer are constant matrices.
Transform Coder and Signal Adapted Filter Bank Optimization
For the special case where the polyphase matrix E(ejω) is paraunitary (ie, unitary for all) and R(ejω) =E†(ejω), the filter bank is called an anorthonormal filter bank. A filterbank whose filters depend on knowledge of the input statistics is called a signal-matched filterbank.
Outline and Scope of the Thesis
- Review of Majorization, Matrix Theory, and Generalized Triangular Decom-
- Transceiver Designs for MIMO Frequency Flat Channels (Chapter 3)
- Transceiver Designs for MIMO Frequency Selective Channels (Chapter 4)
- The Role of GTD in Transform Coding (Chapter 5)
- The Role of GTD in Filter Bank Optimization (Chapter 6)
Other special cases of GTD-TC are GMD (geometric mean decomposition) and BID (bidiagonal transformation). However, the performance of the GMD transform encoder degrades significantly in the low-rate case.
Notations
Additive Majorization and Schur Convexity
Additionally, φ is said to be Schur concave if and only if−φ is Schur convex. Note that the sets of Schur-concave and Schur-convex functions do not constitute a partition of the set of all functions.
Multiplicative Majorization
There are a number of simple but useful facts regarding compositions involving Schur-convex and Schur-concave functions (p.61 of [65]). The following theorem is a direct consequence of the composition rule of Schur convex functions with increasing convex functions.
Relation to Matrix Theory
Hermitian Matrices
Complex-Valued Square Matrices
Generalized Triangular Decomposition
Block-Diagonal Geometric Mean Decomposition
In BD-GMD, one of the identity matrices in (2.4) is restricted to the block diagonal. Third, for the case of linear receivers and transmitters, the joint optimization of the precoder, (linear) equalizer and bit assignment was studied in [48] (under ZF constraint) and in [74] (without ZF constraint).
MIMO Transceivers with Decision Feedback and Bit Loading
Problem Formulation
In the following sections, we first discuss the problem of minimizing the transmitted power subject to a specified total bit rate and a specified error probability at each substream. In this section we will consider the error probability Pe(k) as a quality of service (QoS) specification.
Minimum Power Achieved by DFE Systems
Equality can be achieved in the AM-GM inequality if and only if the terms are identical for allk, i.e. 1 In general (3.15) can give an incomplete or even negative numberbk. However, in the case of high bitrate (largeb), it is large enough to be replaced by integer values without compromising the optimization too much.
GTD-Based Transceivers
With the above choice of transmitter matrices, the error variance (3.5) in the subcurrent becomes 3.8) the transmitted power required to meet the specified QoS constraints and the bit rate can be expressed as. The QR decomposition of the channel matrix can be written as H = QR, where Q has orthonormal columns and is upper triangular.
Other Transceiver Problems Solved by GTD-Based Transceiver
The bit rate maximization problem subject to a transmitted power constraint is the counterpart of the problem described in Eq. Thus, the bit rate maximization problem is reduced to maximizing (3.42) subject to zero forcing.
Simulation Results with Perfect CSI
This means that there exists a GTD for the Hi channel such that both (3.36) and the full bit distribution (3.46) hold simultaneously. Example 1: High bit rate case: In this example we consider GTD transmitters with approximate bit allocation (3.26).
Simulation Results with Limited Feedback
From the plots, we see that the proposed “QR-limited-FB” scheme significantly outperforms the state-of-the-art limited feedback schemes [56, 91] and comes close to the optimal “GMD-perfect-CSI” scheme. Even with such limited feedback, the proposed "QR-limited-FB" scheme works very well.
Concluding Remarks
MIMO Transceivers with Linear Constraints on Transmit Covariance Matrix
- Signal Model and Problem Formulation
- Linear Transceivers
- DFE Transceivers
- Numerical Simulations
- Concluding Remarks
In the first step we will minimize the AM-MSE (arithmetic mean of the mean square error) of the system. In the first step we will minimize the GM-MSE (geometric mean mean square error) of the system.
Conclusions
If the precoder matrix Fi is not square, for example, when the channel matrix is HisN×P and P > M, the rank of U must not be greater than M. When P > M, suppose we first relax the rank constraint, then the rank-relaxing covariance matrixU is solved by the SDP solver as before.
Appendix
Proofs of Lemma 3.2.1
We have shown that the semi-definite programming (SDP) technique provides a nice framework for unifying the design of linear and DFE receivers.
Proofs of Theorem 3.2.4
Therefore, ZP-BD-GMD systems have a much smaller implementation cost than optimal systems. In particular, ZP-BD-GMD systems are average BER minimizers within the family of systems using unitary block diagonal precoders.
Signal Model
In the following sections, we will discuss some important properties of the proposed ZP-BD-GMD receivers. In addition, the receiver structure of the ZP-BD-GMD systems is computationally simple.
Transceivers with Zero-Forcing DFEs
It is therefore important to characterize the performance of the ZF-BD-GMD system for finite block sizes. The following theorem represents the optimality of the ZF-BD-GMD system for any finite block size.
Transceivers with MMSE DFEs
Similar to the ZF case, in the following we derive the BD-GMD system for the MMSE counterpart. We can prove that within this family, the MMSE-BD-GMD system is one of the minimizers for the average BER.
Trade-Off between BW Efficiency and Performance
Since the precoder matrix in the MMSE-BD-GMD system is diagonal block, the transmitter implementation cost is the same as the ZF-BD-GMD case. For the receiver part, we can prove that the implementation cost of the receiver is the same as that of the ZF-BD-GMD system.
ZP for SISO Frequency Selective Channel
Thus, for the example of a frequency-selective SISO channel, systems with a lazy precoder and ZF-DFE are asymptotically optimal in the class of systems with linear precoders and ZF-DFE receivers. In particular, systems with a lazy precoder and MMSE-DFE are asymptotically optimal in the class of systems with a linear unity precoder and an MMSE-DFE receiver.
Numerical Simulations
We can see that the ZF-BD-GMD system indeed has similar performance to the ZF-Optimal system. ZFBDG” represents the ZF-BD-GMD system; "ZFOPT" represents the ZF-Optimal system; and "ZF-Lazy" represents the lazy precoder with zero-forcing DFE.
Concluding Remarks
Appendix
Proof of Lemma 4.3.1
Proof of Theorem 4.3.5
Therefore, to prove this theorem, what we need to prove is that the vector consisting of the absolute values of the subchannel gains of the P0-BD system, the vector consisting of the subchannel gains of the ZF-BD-GMD- system exists, multiplicatively majorizes. Let ap be defined as the product of the largest pabsolute values of the subchannel gains of the P0-BD system, and let bp be defined as the product of the largest psubchannel gains of the ZF-BD-GMD system.
Proof of Theorem 4.3.6
The above advantage of the GMD encoder is shown to be true in the high bit rate case. However, the performance of the GMD transform encoder is degraded in the low-speed case.
GTD Transform Coder for Optimizing Coding Gain
Preliminaries and Reviews
Under the assumption of a high bit rate (5.1), the optimal bit allocation is given by the bit loading formula [35, 107]. At each step of both the Minlab encoder and decoder, a prediction is made based on the quantized data, while in the structure in Fig.
Generalized Triangular Decomposition Transform Coder
Note that this result is true because of the minimum noise structure for the PLT (which has unity noise gain). The first group is the significant group where the K1 data streams contain a rough approximation of the signal.
Simulations
It can be seen from the figure that at the optimal bit load, all GTD-TCs perform approximately the same. It can be seen from the figure that without applied bit load, GMD performs much better than other methods, since GMD without bit allocation is theoretically as good as other methods with optimal bit allocation.
Dithered GMD Transform Coder for Low Rate Applications
Dithered GMD Quantizer
In the GMD-NSD encoder, the quantized signal is directly multiplied by the predictive coefficients for use of subsequent substream quantizers, without first being subtracted from the dither. The complexity of the successive decompositions in the above algorithm is in the same order as that in the GMD transform coder described in [122].
Numerical Example
5.15, we see that in the low-rate regime, the two proposed split transform encoders perform better than all other transform encoders. In the high-rate regime, the two proposed GMD split transform encoders perform comparably to the three encoders ("KLTwBL", "PLTwBL", and "GMD") which are designed under the high-rate assumption.
Concluding Remarks
We will first show that there are two fundamental properties in both the optimal orthonormal GTD FB and the optimal biorthogonal GTD FB, namely, total correlation and spectrum equalization. We will show that the optimal systems are related to the frequency-dependent GTD of the channel response matrix.
Subband Coder Signal Model
The psd matrix of the errore(n) is therefore See(ejω) =R(ejω)Sqq(ejω)R†(ejω)The average mean square error of the coderεcoderis3. 4The optimal bit loading formula for conventional filter banks with perfect reconstruction usually has the granularity problem, that is, the number of bits in the formula must be rounded to the nearest integer when used in practice.
Optimal Orthonormal GTD Filter Banks
If this filter pair also results in the spectrum equalizing property, then the product of the subband variances will be. 6.3.4 we know that the optimal GTD coder will produce the product of the subband variances.
Biorthogonal GTD Filter Banks
For the case where the optimal orthonormal GTD filter banks are designed as GMD filter banks, the diagonal matrix D(ejω) can be written as D(ejω) = Mp. To summarize from Theorem 6.4.1, Corollary 6.4.4 and Corollary 6.4.5, the optimal biorthogonal GTD filter bank also has total decorrelation and spectrum equalization as the two necessary conditions.
Performance Comparison of Optimal Filter Banks Designs
In each case, we also state the necessary and sufficient conditions for optimal solutions. In the following, we compare the coding performance of the optimal subband encoders in these four cases.
The Role of Frequency Dependent GTD in Transceivers for the QoS Problem
Transceivers with Orthonormal Precoder Constraint
We assume at frequencyω, H(ejω) has rankKω, andKω≥M for allω due to the zero-forcing assumption, i.e., there is no channel zero. We are now ready to design our transceiver based on this particular GTD shape of the channel matrix.
Transceivers with Arbitrary Precoder
We have derived the optimal transceiver designs for the case of orthonormal precoder and unconstrained precoder. The results of Theorems 6.6.3 and 6.6.5 provide elegant design methods for both cases – the optimal orthonormal precoder transceiver design can be obtained from a frequency-dependent GTD form of the channel matrix; the optimal unconstrained ZF transceivers can be obtained by first designing the optimal orthonormal transceiver and then cascading with the filter λ(ejω).
Concluding Remarks
This is very similar to the subband coder case, where the optimal biorthogonal GTD subband coders can be obtained by using the optimal orthonormal GTD subband coders and a series of scalar filters.
Appendix
- Proof of Theorem 6.3.1
- Proof of Theorem 6.3.3
- Proof of Lemma 6.6.1
- Proof of Lemma 6.6.2
Vaidyanathan, “Filterbank Optimization with Convex Objectives and Optimality of Principal Component Shapes,” IEEE Trans. Ottersten, “Joint bit allocation and precoding for MIMO systems with decision feedback sensing,” IEEE Trans.
The M -channel maximally decimated filter bank with uniform decimation ratio M
The polyphase representation of the M -channel maximally decimated filter bank
The illustration of the relations between sets of functions [41]
The MIMO transceiver with linear precoder and DFE
The proposed form of optimal solution for the DFE transceiver
The SVD system, which represents a linear transceiver
The QR transceiver, which has the lazy precoder. This is identical to the ZF-VBLAST
Example 1. BER versus Tx-Power for P
Example 2. BER versus Tx-Power for P
BER versus Tx-Power with limited feedback (8 feedback bits per block, and 32 bits
BER versus Tx-Power with limited feedback (8 feedback bits per block, and 24 bits
The system with linear precoding and DFE
A direct implementation of the PLT
The PLT implemented using MINLAB(I) structure
The GTD transform coder implemented using MINLAB(I) structure
The BID Transform coder implemented using MINLAB(I) structure
Use of GTD-TC in the progressive transmission context
Performance of different transform coders with optimal bit allocation. Input covari-
Performance of different transform coders with optimal bit allocation. Input covari-
Comparison of coding gain of different transform coders with optimal bit allocation
Performance of different transform coders with uniform bit allocation. Input covari-
Performance of different transform coders with uniform bit allocation. Input covari-
Subtractive dithered GMD transform coder
Nonsubtractive dithered GMD transform coder
The equivalent model of dithered GMD transform coder
Performance of different transform coders
The biorthogonal GTD subband coders for M = 4
A restricted case of the biorthogonal GTD subband coders for M = 4
Coding gain of subband coders with M = 3 for the AR(1) process with ρ from 0.85 to
Coding gain of subband coders with M = 4 for the AR(2) process with ρ from 0.95 to
Monotone behavior of the coding gain as a function of the number of channels for the
Nonmonotone behavior of the coding gain as a function of the number of channels for
Schematic of a frequency selective transceiver with linear precoder and zero-forcing