Decomposed real and imaginary balancer

Abstract

Disclosed are split real and imaginary equalizers. In an example of a signal equalizer described, a first filter stage is configured to perform adaptive equalization of crosstalk between a first signal component and a second signal component of a composite signal (1502). A second filter stage is coupled in series to the first filter stage. The second equalizer stage is configured to perform separate adaptive equalization of the first signal component and separate adaptive equalization of the second signal component (1506).

Description

Technical Field

[0001] This application relates to an adaptive equalizer that is decomposed into separate real and imaginary stages. Background Technology

[0002] Based on the concept of real numbers, a complex number is a number of the form a + bi, where a and b are real numbers, and i is an indefinite number, satisfying i 2 = -1. The real number a is called the real part of the complex number a + bi; the real number b is called its imaginary part.

[0003] For dielectric waveguide cables in the millimeter-wave band (110-140 GHz), the channel group delay is typically an asymmetric parabola. Therefore, the channel impulse response is a composite multipath response, with multipath crosstalk existing between the transmitted in-phase and quadrature-phase components. For passband transmission, two independent signals can be transmitted by modulating the amplitude and phase of a sinusoidal signal. Equivalently, two independent signals can be transmitted by modulating the amplitudes of two sinusoidal signals with a 90-degree phase shift. The equivalent system is modeled as a composite baseband (in-phase and quadrature-phase) transmit / receive signal. For example, QAM (Quadrature Amplitude Modulation) or QPSK (Quadrature Phase Shift Keying) transmission methods can be used. When a passband signal (or its equivalent composite baseband signal) is transmitted through a composite channel (where each path has a different phase, such as a DWG channel), the two transmitted independent components, the amplitude and phase of the passband signal (equivalent to the in-phase and quadrature-phase of the composite baseband signal), interfere with each other. Therefore, in addition to the multipath channel effects on each component, they will also suffer from crosstalk between them.

[0004] Dielectric waveguides (DWGs) are high-frequency alternatives to copper wires and optical fibers. Dielectric waveguides use a solid dielectric core instead of a hollow tube. A dielectric is an electrically insulating material that can be polarized by an applied electric field. When a dielectric is placed in an electric field, charges do not flow through the material as they would in a conductor, but rather shift slightly from their average equilibrium position, resulting in dielectric polarization. Due to dielectric polarization, positive charges shift toward the electric field, and negative charges shift in the opposite direction. This creates an internal electric field, which weakens the overall electric field within the dielectric itself. If the dielectric is composed of weakly bonded molecules, these molecules will polarize and reorient themselves, so that their axes of symmetry align with the field. The term "insulator" indicates low conductivity, while "dielectric" is generally used to describe materials with high polarizability; high polarizability is represented by a number called the dielectric constant (εk). The term insulator generally indicates an electrical barrier, while the term dielectric indicates the material's energy storage capacity through polarization.

[0005] A single-stage adaptive composite equalizer with complex taps for in-phase and quadrature phase components helps to reverse signal distortion as the signal passes through a channel (e.g., through a DWG). The equalizer can be a linear composite equalizer with only a feedforward (FF) filter, or a decision feedback (DFE) composite equalizer with both feedforward and feedback (FB) sections. Summary of the Invention

[0006] In one example of a described signal equalizer, a first filter stage is configured to perform adaptive equalization of crosstalk between a first signal component and a second signal component of a composite signal. A second filter stage is coupled in series to the first filter stage. The second equalizer stage is configured to perform individual adaptive equalization of the first signal component and the second signal component. Attached Figure Description

[0007] Figure 1A and Figure 1B This is a block diagram illustrating a system with a composite communication channel.

[0008] Figure 2A and Figure 2B This is a graph showing the group delay of the signal propagated through DWG.

[0009] Figure 3 This is a block diagram of an example adaptive equalizer for composite channels.

[0010] Figures 4 and 5 illustrate conventional feedforward filters used in composite channels.

[0011] Figures 6 to 10 This is a block diagram of an example two-stage filter for decomposing composite channels.

[0012] Figure 11 This is a block diagram of a composite channel filtering model.

[0013] Figure 12 This is an example implementation of an equalizer for composite channels that uses standard SERDES circuitry to form a decomposed two-stage filter.

[0014] Figure 13 This is a block diagram of an example technique for joint adaptation of filter structures used for decomposition.

[0015] Figure 14 This is a block diagram of an example technique for independent adaptation of filter structures used for decomposition.

[0016] Figure 15 This is the operation flowchart of a two-stage filter used for decomposing composite channels. Detailed Implementation

[0017] In the accompanying drawings, similar elements are indicated by similar reference numerals for consistency.

[0018] For dielectric waveguide (DWG) cables in the millimeter-wave band (110-140 GHz), the channel group delay is typically asymmetric parabolic. Therefore, the channel impulse response is a complex multipath response, generating multipath crosstalk between the in-phase and quadrature-phase components of the transmitted signal. Generally, a single-stage adaptive composite equalizer with composite taps for both in-phase and quadrature-phase components is useful for addressing distortions that occur when inverted signals traverse the channel (e.g., through a DWG). However, equalizers with composite taps for operation in the millimeter-wave band are difficult to implement and manufacture in a single integrated circuit.

[0019] Adaptive equalizer architectures provide suboptimal equalization for time-varying characteristics of the propagation channel. However, these equalizers are attractive because their computational complexity is lower than that of maximum likelihood sequence estimation (MLSE) equalizers. Adaptive equalizer architecture options include linear and decision feedback. Adaptive algorithm options include least mean square (LMS), recursive least squares (RLS), and constant modulus (CMA) algorithms. The general operation of adaptive equalizers is described, for example, in the "Adaptive Equalizers" section of Mathworks Release Notes R2019b, 2019.

[0020] To decode the received signal, the adaptive equalizer first applies a finite impulse response (FIR) feedforward filter to the symbols in the input signal. The FIR filter tap weights correspond to the channel estimate. The adaptive filter then outputs a signal estimate, which is used to update the tap weights of the next symbol using a feedback filter. The signal estimate and weight update depend on the structure and algorithm of the adaptive equalizer.

[0021] In the example described in this paper, a decomposed cascaded two-stage feedforward filter replaces a single-stage composite feedforward filter. The decomposed two-stage filter has one stage performing pure real equalization, while the other stage performs pure virtual equalization. The example decomposed two-stage filter structure described in this paper offers lower complexity, greater flexibility, and better performance than a conventional single-stage composite equalizer. In various examples, the feedback filter portion of the adaptive equalizer can be single-stage, decomposed, or completely removed.

[0022] In the described example, the low-complexity simulation implementation reduces the load on critical signals and simplifies delay matching between real and virtual paths.

[0023] In some examples, the first stage is configured as a purely virtual stage. This allows the second stage to use the standard serializer / deserializer (SERDES) module.

[0024] As described in more detail in this paper, flexible structures have different complexity / performance trade-offs.

[0025] Figure 1A This is a block diagram of an example system 100, which includes a QAM transmitter 102, a QAM receiver 103, and a composite communication channel 104 coupled between the output of the QAM transmitter 102 and the input of the QAM receiver 103. In this example, the QAM transmitter 102 generates a composite signal comprising an in-phase signal component 105 and a quadrature-phase signal component 106. The receiver 103 receives the in-phase composite signal component 111 and the quadrature-phase composite signal component 112, equalizes the composite signal, and then demodulates the composite signal using known or later-developed demodulation techniques. The demodulated signal is then output for further processing by an associated system (not shown) using known or later-developed communication and / or signal processing techniques.

[0026] The DWG 108 exhibits multipath channel effects. Therefore, if information is modulated in both the amplitude and phase of the transmitted passband sinusoidal signal, the DWG 104 will affect both amplitude and phase simultaneously. Equivalently, in the baseband equivalent model, the in-phase and quadrature-phase components of the information signal will be affected due to the effective multipath channel. The QAM receiver 103 includes an equalizer, described in more detail below, to mitigate the effects of multipath channel effects in order to improve the signal-to-noise ratio (SNR).

[0027] Figure 1A A composite communication channel 104 as a passband model is shown. In this example, mixer 107 represents the upconversion of the transmitted signal from a composite baseband to a real passband signal. Adder 109 represents the effect of ambient noise added to the composite signal during transmission through DWG 104. Mixer 110 represents the further downconversion of the received real passband signal to an equivalent composite baseband signal. In this example, the noise is assumed to be additive white Gaussian noise (AWGN).

[0028] Figure 1B This is a block diagram of a composite communication channel 104 as an equivalent model of a composite baseband. In this example, transmission medium 121 transmits a composite signal including an in-phase signal component 105 and a quadrature phase component 106. A composite AWGN (with an in-phase component 123 and a quadrature phase component) is injected into the composite signal via adder 122.

[0029] Figure 2A and Figure 2B This is a graph of the group delay of the signal propagating through the example DWG. Figure 2A The figure shows a graph of group delay (ps) versus frequency (GHz) for 200 groups in the 50-200 GHz range. Figure 2BThe figure shows an expanded view of the curves for two different simulation models in the 110-140 GHz range, represented by graphs 201 and 202. Initial measurements show that the DWG channel in the 110-140 GHz band has an asymmetric parabolic group delay, which leads to multipath interference as well as multipath crosstalk between in-phase and quadrature-phase transmitted data. For example, for a passband from 110 to 140 GHz, the group delay variation from the lower edge of the band (110 GHz) to the center of the band (125 GHz) is approximately 9 ps / m, and the group delay variation from the center of the band to the upper edge (140 GHz) is approximately 6 ps / m. Furthermore, if the group delay is not centered at the center frequency (125 GHz), the equivalent group delay becomes even more asymmetric.

[0030] Figure 3 This is a block diagram of an example SERDES equalizer 300 for composite channels operating at 100 GHz or higher. The equalizer 100 includes a composite feedforward filter 301, a composite feedback filter 304, and a decision module 306. The decision module 306 generates an output signal 307, which is an equalization estimate of the input signal obtained by subtracting the output of the feedback filter from the output of the feedforward filter using a subtractor 305. In this example, the input signal is a QPSK signal. Other examples can be designed in a similar manner for other orthogonal composite signals. As described below, the composite feedforward filter 301 is a decomposed cascaded two-stage feedforward filter. In some examples, the decision feedback filter 304 is a decomposed cascaded two-stage feedback filter.

[0031] Figure 4 is a block diagram of a conventional single-stage composite feedforward filter 401. The filter module w is used for the in-phase portion of the channel. I 410 and filter module w for the orthogonal phase portion of the channel Q 411 is connected to the input for the in-phase signal. Similarly, the filter module w is used for the quadrature phase portion of the channel. Q 412 and filter module w for the in-phase portion of the channel I 413 is connected to the input for the quadrature phase signal. Adder 414 combines the outputs of in-phase filter 410 and quadrature phase filter 411 to form an output for the in-phase transmission component, while adder 414 combines the outputs of in-phase filter 413 and quadrature phase filter 412 to form an output for the quadrature phase transmission component.

[0032] Figure 5 is a more detailed schematic diagram of the single-stage feedforward filter 401. The in-phase filter 410 includes a digital delay line 520 that delays the in-phase output signal I_OUT provided by comparator 523 by one clock cycle at each tap. A multiplication unit 521 multiplies each tap coefficient value by the data provided by the delay line 520. The outputs of each stage of the multiplication unit 521 are combined on a differential adder bus 522, which is sequentially coupled to the input of comparator 523. A differential input receiver 526 is coupled to the differential summing bus 522.

[0033] The quadrature phase filter 412 operates in a similar manner, with the outputs of the multiplication unit 525 combined on the differential adder bus 525. The adder bus 525 is coupled to the adder bus 522, thereby adding the output of the in-phase filter 410 to the output of the quadrature phase filter 412, as shown in adder 414 (Figure 4).

[0034] In a similar manner, comparator 527 forms the quadrature phase output signal Q_OUT based on the state of the differential adder bus 529, which is coupled to the outputs of the quadrature phase filter 411, the in-phase filter 413, and the differential input receiver 528.

[0035] One problem with this method is the load generated on the corresponding multiplication units within the differential input receivers 526, 528 and filters 410-414. For conventional semiconductor manufacturing techniques, reliable operation at a target data rate of 100Gbps using the 110-140GHz band is not feasible with the feedforward filter 401 configuration.

[0036] Figure 6 It is used for Figure 3 A block diagram of a two-stage filter 301 for example decomposition of the composite channel within the adaptive equalizer 300 is shown. As mentioned above, an analog implementation using a single-stage composite filter (shown as filter 401 in Figures 4-5) to support a target data rate of 100Gbps is too complex for conventional semiconductor manufacturing techniques to produce integrated circuits at an acceptable commercial price. Similarly, a digital implementation using digital signal processing techniques is not feasible at 100Gbps.

[0037] One option might be to omit the composite filtering and perform only real filtering, ignoring the composite crosstalk effects. However, simulations have shown that, due to high crosstalk, a simple all-real filter equalizer is insufficient to achieve a low bit error rate in the target DWG environment. For example, the target bit error rate for the example system could be less than 1e-12.

[0038] To mitigate the effects of composite channels, the example described in this paper decomposes the single-stage composite equalizer feedforward filter into two cascaded stages, such as a first stage 602 and a second stage 603 coupled in a series configuration, as follows: Figure 6 As shown. One stage performs real filtering, and the other performs virtual filtering. The filtering operation of the two-stage decomposition structure is equivalent to a single-stage composite filter with a comparable number of taps. In various examples, the LMS (Least Mean Square) adaptation of the structure used for decomposition can be implemented as a joint or independent adaptation for both stages, as described in more detail below. The number of taps in each stage can typically be different, which adds considerable flexibility to the decomposition structure. Furthermore, example two-stage adaptive composite equalizers can be selectively designed to perform either real or virtual filter functions first.

[0039] The feedback filter used in the example equalizer is less complex than the feedforward filter because the number of taps and input stages in the feedback filter can be smaller than those used in the feedforward filter. Therefore, the example feedback filter can be implemented as a composite single-stage filter, or it can be implemented as a two-stage decomposition filter in a similar manner to the feedforward filter.

[0040] Still referencing Figure 6 The first stage 602 of the feedforward filter 301 uses tapped delay line quadrature phase filters 611 and 612 to perform adaptive filtering on the crosstalk between the in-phase signal component received at input 621 and the quadrature phase signal component received at input 622. Adder 614 subtracts the crosstalk correction output of filter 612 from the in-phase signal 621 and outputs a crosstalk-compensated in-phase signal 621. Similarly, adder 615 adds the crosstalk correction output of filter 611 to the quadrature phase signal 622 and outputs a crosstalk-compensated quadrature phase signal 624. The second stage 603 uses in-phase filter 610 to perform separate adaptive filtering on the compensated in-phase signal component 623 to generate a filtered in-phase signal 625. Simultaneously, the second-stage filter 603 uses in-phase filter 613 to perform separate adaptive filtering on the compensated quadrature phase signal component 624 to generate a filtered quadrature phase signal 626.

[0041] In this example, the quadrature phase filters 611 and 612 are configured to conform to expression (1) in order to maintain w' Q Followed by w I The cascaded system and the quadrature phase filter in Figure 4 (w Q The equivalence of 411 and 412.

[0042] w Q (n)=w′ Q (n)*w I (n) (1)

[0043] The asterisk (*) indicates a convolution operation.

[0044] Figure 7 It is applicable Figure 6A schematic diagram of an example tapped delay line filter 710 of filters 611-614 is provided. Example filter 710 includes a digital delay line 720 that delays the decision signal 727 provided by comparator 723 by one clock cycle at each tap. Multiplication unit 721 multiplies each tap coefficient value by the data (0 or 1) provided by delay line 720. The outputs of each stage of multiplication unit 721 are combined on a differential adder bus 722, which is coupled in turn to the input of comparator 723. Differential input receiver 726 is coupled to differential summing bus 722.

[0045] In example filter 710, the number of taps in delay line 720 and the corresponding number of multiplier stages 721 can be selected to optimize the filter design for a chosen operating frequency. Therefore, in example filter 301, each of filters 611-614 can have a different number of taps. In another example filter 301, for example, filters 611-614 can all be implemented with the same number of taps.

[0046] Figure 8 This is another example of a two-stage filter 801 used for decomposing composite channels. A two-stage filter 801 can be used instead... Figure 3 The two-stage filter 301 is described. In this example, the first stage 802 of the feedforward filter 801 uses tapped delay line in-phase filters 810 and 813 to perform adaptive filtering on crosstalk between the in-phase signal component received at input 821 and the quadrature-phase signal component received at input 822. Adder 814 subtracts the quadrature-phase signal 822 from the crosstalk-corrected output of filter 810 and outputs a crosstalk-compensated in-phase signal 823. Similarly, adder 815 adds the crosstalk-corrected output of filter 813 to the in-phase signal 821 and outputs a crosstalk-compensated quadrature-phase signal 824.

[0047] The second stage 803 is coupled in series to the first stage 802 and uses a quadrature phase filter 811 to perform separate adaptive filtering on the compensated in-phase signal component 823 to generate a filtered in-phase signal 825. Simultaneously, the second-stage filter 803 uses a quadrature phase filter 812 to perform separate adaptive filtering on the compensated quadrature phase signal component 824 to generate a filtered quadrature phase signal 826.

[0048] In this example, in-phase filters 810 and 813 are configured to conform to expression (2) in order to maintain w' I Followed by w Q The cascaded system and the in-phase filter in Figure 4 (w Q The equivalence of 410 and 413.

[0049] w I(n)=w′ I (n)*w Q (n) (2)

[0050] The asterisk (*) indicates a convolution operation.

[0051] Figure 9 This is another example of a two-stage filter 901 used for decomposing composite channels. A two-stage filter 901 can be used instead... Figure 3 The two-stage filter 301 in the example. In this example, the operation sequence of the first stage 902 and the second stage 903 is as follows: Figure 6 Example second-order filter 601 and Figure 8 The filter 801 is the opposite. In this example two-stage filter 901, the first stage 902 is configured to perform: separate adaptive equalization of a first signal component of the composite signal received at input 921, and separate adaptive equalization of a second signal component of the composite signal received at input 922. The second stage 903 is coupled in series to the first stage 902 and is configured to perform adaptive equalization of crosstalk between the first and second signal components of the composite signal transmitted through the composite channel.

[0052] In this example, the first stage 902 uses an in-phase filter 910 to perform individual adaptive filtering on the received in-phase signal component 921 to produce a filtered in-phase signal 923. Simultaneously, the first-stage filter 902 uses a quadrature-phase filter 913 to perform individual adaptive filtering on the received quadrature-phase signal component 922 to produce a filtered quadrature-phase signal 924. The second stage 903 of the feedforward filter 901 uses tapped delay line quadrature-phase filters 911 and 912 to adaptively filter the crosstalk between the filtered in-phase signal component 923 and the filtered quadrature-phase signal component 924. Adder 914 subtracts the crosstalk correction output of filter 911 from the filtered in-phase signal 923 and outputs a composite filtered in-phase signal 925. Similarly, adder 915 adds the crosstalk correction output of filter 912 to the filtered quadrature-phase signal 924 and outputs a composite filtered quadrature-phase signal 926.

[0053] In this example, the quadrature phase filters 911 and 912 are configured to conform to expression (1) in order to maintain equivalence with the quadrature phase filters 411 and 412 of FIG4.

[0054] Figure 10 This is another example of a two-stage filter 1001 used for decomposing composite channels. A two-stage filter 1001 can be used instead... Figure 3 The two-stage filter 301 in the example. In this example, the operation order of the first stage 1002 and the second stage 1003 is... Figure 6 Example second-order filter 601 and Figure 8 The filter 801 is the opposite.

[0055] In this example, the first stage 1002 uses a quadrature phase filter 1011 to perform individual adaptive filtering on the received in-phase signal component 1021 to generate a filtered in-phase signal 1023. Simultaneously, the first-stage filter 1002 uses a quadrature phase filter 1012 to perform individual adaptive filtering on the received quadrature phase signal component 1022 to generate a filtered quadrature phase signal 1024. A second stage 1003 of the feedforward filter 1001 is coupled in series to the first stage 1002 and uses tapped delay line in-phase filters 1010 and 1013 to adaptively filter the crosstalk between the filtered in-phase signal component 1023 and the filtered quadrature phase signal component 1024. An adder 1014 subtracts the filtered quadrature phase signal 1024 from the output of the in-phase filter 1010 and outputs a composite filtered in-phase signal 1025. Similarly, adder 1015 adds the output of quadrature phase filter 1013 to the filtered in-phase signal 1023 and outputs a composite filtered quadrature phase signal 1026.

[0056] In this example, in-phase filters 1010 and 1013 are configured to conform to expression (2) in order to maintain equivalence with the quadrature phase filters 410 and 413 of FIG4.

[0057] Equivalence analysis

[0058] It can be proven using linear systems theory that each of Examples 301, 801, 901, and 1001 produces an equivalent filtering function. For example, all filters (w) I w' I w Q w' Q The system is linear, and the adders (614 and 615) in design 301 can be moved to the end of the design, i.e., after 610 and 613, which will result in two units with w. I (610) branch and two with cascaded w' Q (611) and w I (610) branch. Furthermore, for cascaded linear systems, for example, w' Q (611) and w I (610) Filters can be reversed in order and have exactly the same function. This would make design 301 equivalent to design 901. By comparison, given expressions (1) and (2), design 301 is equivalent to design 801, and design 901 is equivalent to design 1001. Therefore, all four designs 301, 801, 901, and 1001 are equivalent.

[0059] Figure 11This is a block diagram of the composite channel model 1101. (Reference) Figure 1A and Figure 1B Communication channel 104 can be modeled as a linear system (linear filtering). In this example, C I 1110 and 1113 represent the real part of the composite channel, and C Q 1111 and 1112 represent the imaginary part of the composite channel. C Q 1111 and 1112 are the causes Figure 1A In-phase (T) I )105 and Figure 1A orthogonal phase (T) Q The channel portion is affected by crosstalk interference between transmitted signals. If the transmitted signals are real PAM (not composite QAM), then the composite channel will not have a problem because C Q It will only leak and can be ignored.

[0060] Similarly, composite channel 104 ( Figure 1B ) can be modeled using matrix representation, because linear filtering can be represented using Toeplitz matrix multiplication, as shown in expression (3).

[0061]

[0062] In the case of a multi-tap channel, C I and C Q It is a Toeplitz convolution matrix. Furthermore, T... I T Q R I R Q n I and n Q It is a vector. For the special case where the channel is single-tap, all elements in the matrix of expression (3) will be scalars. If C Q If the value is zero, the channel matrix will be diagonal and no crosstalk will occur.

[0063] Therefore, the equivalence between Design 301 and 901 can be analyzed using matrix representation. Consider R = [R I ;R Q ] as the received input vector, and the first level 602 ( Figure 6 The instruction is A and the second level is 603 ( Figure 6 The indication is B. In this case, filter 301 ( Figure 6 Output vector It can be represented as Matrix multiplication. Similarly, filter 901 ( Figure 9 Output vector It can be represented as Filter design 301 is equivalent to filter design 901, therefore If it can be proven that matrix multiplication AB is commutative, i.e., AB = BA. However, for matrices, multiplication is generally not commutative. Some special cases, including multiplication of two diagonal matrices, or multiplication with the identity matrix, are commutative.

[0064] Another special case is when both matrices are diagonalizable, matrix multiplication will also involve commutation. This occurs if there exists a matrix T such that both matrices can be factored into TD. a T -1 and TD b T -1 Then A and B can be diagonalized simultaneously, where D a and D b It is a diagonal matrix. This is also known as matrices A and B being diagonalizable on the same basis.

[0065] Asymptotically interchangeable Toeplitz matrices (as the row and column dimensions approach infinity). This means they are diagonalized on the same basis. When the dimensions of the input vector and the Toeplitz matrix approach infinity, linear convolution can be represented as the product of the input data vector and the Toeplitz matrix.

[0066] If all filters have only one scalar coefficient, a special case arises where each filter is simply a scalar multiplication by (w). I w′ I w Q w′ Q Therefore, it is clear that the two classes A and B are commutative. The reason is that class A (e.g., 603, 902) can be written as (w I I), where w I I is a scalar coefficient and I is the identity matrix. Therefore, as shown in expression (4), level A can be interchanged with any matrix.

[0067]

[0068]

[0069] For a multi-coefficient filter w' Q and w I In expression (4), levels A and B will be block matrices, where filters 610, 611, 612, and 613 are represented as Toeplitz matrices. Filter design 301 can be represented as shown in expression (5).

[0070]

[0071]

[0072] Among them W' Q W I It is a Toeplitz convolution matrix, and S is of the form The switching matrix. Operators This represents Kronecker multiplication.

[0073] In the final step of expression (5), the mixed product property of Kronecker multiplication is used: if X, Y, Z, and V are matrices such that matrix products XZ and YV can be formed, then

[0074] Similarly, the system model of filter design 901 can be represented as shown in expression (6).

[0075]

[0076]

[0077] W' Q and W I It is asymptotically commutative, that is, W I W′ Q =W′ Q W I Since they are both Toeplitz matrices, expressions (5) and (6) are equivalent, i.e., Therefore, filter design 301 is equivalent to filter design 901.

[0078] Figure 12 This is an example implementation of a two-stage filter 1201 for composite channels, using standard SERDES modules 1231 and 1232 to form a decomposed two-stage filter. Typically, a serializer-deserializer (SERDES) module performs serial-to-parallel conversion on data received from a peripheral device and parallel-to-serial conversion on data received from a CPU (Central Processing Unit). SERDES modules may include control capabilities and processor interrupt systems, which can be customized to, for example, minimize the software management of the communication link. Design libraries may include conventional SERDES modules for more complex design projects. Standard SERDES modules may include real-only equalizers to provide signal equalization for non-composite communication signals.

[0079] In this example, tapped delay line quadrature phase filters 1211 and 1212 are interconnected in a manner similar to... Figure 6 The tapped delay line quadrature phase filters 611 and 612 operate in a manner similar to those in the original text, to form a shape similar to... Figure 6The first stage 1202 operates in the manner described in the first stage 602. The second stage 1203 is implemented using two standard real-only SERDES modules 1231 and 1232 to perform real filtering only on the real components of the in-phase signal 1221 and the quadrature-phase signal 1222 received at the inputs of the two-stage filter 1201. The second stage 1203 uses the in-phase filter within the real-only SERDES module 1231 to perform individual adaptive filtering on the compensated in-phase signal component 1223 to produce the filtered in-phase signal 625. Simultaneously, the second-stage filter 603 uses the in-phase filter within the real-only SERDES module 1232 to perform individual adaptive filtering on the compensated quadrature-phase signal component 1224 to produce the filtered quadrature-phase signal 1226. The real-only SERDES modules 1231 and 1232 provide similar functionality to... Figure 6 The filtering function of the in-phase filters 610 and 613.

[0080] adapt

[0081] The filtering operation on the decomposed structure is equivalent to the single-stage composite filtering described above. However, LMS adaptation is different for the decomposed structure and can be performed in a different manner. LMS adaptation is used to simulate the desired filter by finding the filter coefficients that have the least mean square correlation with the signal generating the error. For example, adaptation adjusts the corresponding filter coefficients for each tap, such as in... Figure 7 In the multiplication unit 721.

[0082] Figure 13 This is a block diagram of an example technique for the joint adaptation of the decomposed filter structure 1301 within equalizer 1300 to a composite channel operating at 100 GHz or higher. The joint adaptation minimizes the final error 1341 in the output signal 1307 of equalizer 1300. Equalizer 1300 includes a composite feedforward filter 1301, a composite feedback filter 1304, and a decision module 1306. The decision module generates the output signal 1307 by subtracting the output of the feedback filter 1304 from the output of the feedforward filter using a subtractor 1305. This output signal 1307 is an equalization estimate of the input signal 1321. The two-stage feedforward filter 1301 is decomposed into two stages 1302 and 1303, which can be similar to any of the two-stage filters 301, 801, 901, and 1001 described in more detail above.

[0083] Joint adaptation yields optimal performance, similar to adaptation of a single-stage composite filter. However, it requires greater complexity because each filter 1302, 1303 must be adapted using the output of another filter.

[0084] In this example, the error signal e 1341 is generated by subtractor 1342 according to expression (7). w1, w2 and wfb These are vectors for filter stage 1, filter stage 2, and feedback filter, respectively. The vectors r1 and r2 in the error signal calculation are represented by expressions (8) and (9), respectively.

[0085] e = s - (w²r¹ - w) fb s)=s-(w1r2-w fb s) (7)

[0086] r1(n)=w1(n)-r(n) (8)

[0087] r2(n)=w2(n)*r(n) (9)

[0088]

[0089]

[0090] In expressions (10) and (11), sgn() represents symbolic operation.

[0091]

[0092] Using sgn(r1*), sgn(r2*), and sgn(e) in the LMS adaptation expressions (10) and (11) is a suboptimal approach to achieving a lower complexity design. However, for a preferred adaptation, the sgn() operation may be omitted and r1*, r2*, and e may be used directly.

[0093] In expression (10), w1 represents the first filter stage 1302 and is either a purely real filter or a purely virtual filter. The same applies to w2, which represents the second filter stage 1303 in expression (11). Therefore, in the examples of real w1 and virtual w2, r1 is the output of the purely real filter w1 used to modify the purely virtual filter w2. The same applies to r2.

[0094] For optimal joint adaptation, both r1 and r2 are required. However, in the main data path, only r1 is available. Therefore, for optimal adaptation, filter 2 is implemented twice, in 1303 for the main data path and in 1344 for the adaptation path. To avoid designing the same filter (filter 2) twice, the received signal r1321 can be used to approximate both r1 and r2 in the adaptation algorithm shown by expressions (10) and (11), respectively. Simulations show that this is a good approximation.

[0095] Figure 14This is a block diagram of an example technique for the independent adaptation of the decomposed filter structure 1401 within equalizer 1400 to a composite channel operating at 100 GHz or higher. Equalizer 1400 includes a composite feedforward filter 1401, a composite feedback filter 1404, and a decision module 1406 that generates an output signal S 1407, which is an equalization estimate of the input signal 1421, after subtracting the output of the feedback filter 1404 from the output of the feedforward filter 1401 using a subtractor 1405. The two-stage feedforward filter 1401 is decomposed into two stages 1402 and 1403, which can be similar to any of the two-stage filters 301, 801, 901, and 1001 described in more detail above.

[0096] Independent adaptation is suboptimal because the adaptation is performed independently. However, independent adaptation is not very complex because each stage is completely independent. Since the adaptation is independent, the standard SERDES module can be used for the second-stage filter 1403, similar to... Figure 12 SERDES1231, 1232.

[0097] In this example, the error signal e1 1441 is generated by subtractor 1442 according to expression (12). Decision module 1456 generates an output signal S1 as an estimate of the intermediate signal r1. The error signal e2 1451 is generated by subtractor 1452 according to expression (13).

[0098] In expression (15), w1 represents the first filter stage 1402 and is either a purely real filter or a purely virtual filter. The same applies to w2, which represents the second filter stage 1403 in expression (16). Therefore, in the case of real w1 and virtual w2, r1 is the output of the purely real filter w1, which is used to independently adapt the purely virtual filter w2.

[0099] e1 = s - (w2r1 - w) fb s) (12)

[0100] e2=s1-(w1r) (13)

[0101] r1(n)=w1(n)*r(n) (14)

[0102]

[0103]

[0104] LMS adaptation module 1453 receives error signal e2 and generates an output signal, which is coupled to control the adaptation of the filter coefficients of the first filter stage 1402 using known or later-developed LMS adaptation techniques. LMS adaptation module 1443 receives error signal e1 and generates an output signal, which is coupled to independently control the adaptation of the filter coefficients of the second filter stage 1403 using known or later-developed LMS adaptation techniques.

[0105] Optimize the decision-making process

[0106] The decision taps can be optimized during runtime without additional hardware. For decision-oriented modes, changing the filter initialization is sufficient to change the decision taps. For example, refer again... Figure 7 The tap delay line 720 can be initialized to all zeros, except when the decision tap is set to 1 at runtime. This allows system designers to try all possible initializations and select the optimal decision tap based on the average equalizer error signal.

[0107] Table 1 summarizes the simulated signal-to-interference ratio (SIR) for various equalizer configurations, including symmetrical, Rosenberger, asymmetrical 7 / 5, and asymmetrical 7 / 5 with a frequency offset of +5%. The first column of the results is for a conventional single-stage equalizer, as shown in Figure 5. The second column is for a feedforward filter with joint adaptation decomposition, such as... Figure 13 As shown. The third column refers to feedforward filters with independent adaptive decomposition, where the first stage filters the imaginary component of the composite signal, while the second stage filters the real component, as shown. Figure 14 As shown.

[0108] Table 1 - LMS Decomposition Filter Equalizer with 5-Tap Composite Feedforward Filter and 1-Tap Composite Feedback Filter

[0109]

[0110] As shown in Table 1, structures with joint adaptive decomposition achieve nearly the same performance as single-stage composite structures. For structures with independent adaptive decomposition, the virtual-real performance shows some degradation because it is a suboptimal approach, as previously described.

[0111] Figure 15 This is a flowchart illustrating the operation of a two-stage filter used for decomposing a composite channel. The two-stage feedforward is decomposed into two stages, which can be similar to any of the two-stage filters 301, 801, 901, and 1001 described in more detail above.

[0112] At position 1502, a composite signal is received from the composite communication channel. In various examples, the composite signal includes in-phase and quadrature-phase components, such as those generated by a QAM transmitter. In other examples, the composite signal may be generated by a QPSK transmitter. In still other examples, other known or subsequently developed techniques can be used to generate the composite communication signal. In this example, the composite communication signal propagates through a DWG with multipath channel effects. Therefore, if information is modulated in the amplitude and phase of the transmitted passband sinusoidal signal, the DWG will affect both amplitude and phase. Equivalently, in the baseband equivalent model, the in-phase and quadrature-phase components of the information signal will be affected.

[0113] At 1504, one of the filter stages is configured to perform adaptive equalization of crosstalk between the first and second signal components of the composite signal, as described in more detail above, for example, for... Figure 6 Intermediate 602 Figure 8 802 in Figure 9 903 or Figure 10 As described in 1003.

[0114] At 1506, another filter stage is configured to perform separate adaptive equalization for the first signal component and separate adaptive equalization for the second signal component, as described in more detail above, for example, for... Figure 6 Intermediate Level 603 Figure 8 803 in Figure 9 902 or Figure 10 As described in 1002.

[0115] In some examples, crosstalk equalization is performed first, followed by individual signal component equalization. In other examples, crosstalk equalization is performed second, while individual signal component equalization is performed first.

[0116] In some examples, independent adaptation of two load balancers is performed at 1508. Examples of independent adaptation are described in more detail above, such as for... Figure 14 The decomposition filter 1401 is described in the text.

[0117] In some examples, a joint adaptation of two equilibrium levels is performed at 1510. Examples of joint adaptation are described in more detail above, such as for... Figure 13 The decomposition filter 1301 in the process.

[0118] At 1512, a sample of the equalized composite signal is output for further processing in an associated signal processing system coupled to receive the output signal. The process at 1502 is repeated continuously as each sample or portion of the output signal is output.

[0119] In this way, for DWG composite channels (or any composite channel), several advantages are offered through the lower complexity and greater flexibility of decomposing the two-stage cascaded real-virtual or virtual-real filter structure. The decomposed filter structure offers flexibility in design choices for joint or independent LMS adaptation of the two stages. At 1500, the system designer can choose whether the number of taps in each stage is the same or different. The position of each stage (i.e., which is the first and which is the second) can vary.

[0120] Depending on system accuracy and cost considerations, feedback filters can be designed as single-stage composite feedback or decomposed feedback filters.

[0121] The example can be implemented as a standalone solution using joint adaptation, which outperforms single-stage solutions and has lower complexity, or integrated with current SERDES solutions by setting the first stage as a purely virtual filter that is modified independently.

[0122] Other embodiments

[0123] In the described example, the received composite signal is a QPSK type composite signal. In other examples, various types of composite signals with orthogonality or other phase shifts can be accommodated by appropriately selecting the taps and tap coefficients.

[0124] In the described example, the feedback filter is a single-stage composite filter. In another example, the feedback filter can be designed as a decomposed two-stage filter.

[0125] The example described illustrates a decision feedback composite equalizer. In another example, a linear composite equalizer (which only has a feedforward filter) can be implemented.

[0126] The example described depicts a composite signal with in-phase and quadrature phase components, thus exhibiting a 90-degree phase relationship. In another example, different phase relationships may exist, such as a 120-degree relationship, a 45-degree relationship, etc.

[0127] In this specification, the term "coupled" and its derivatives refer to indirect, direct, optical, and / or wireless electrical connections. Thus, if a first device is coupled to a second device, the connection can be a direct electrical connection, an indirect electrical connection via other devices and connections, an optoelectronic connection, and / or a wireless connection.

[0128] Within the scope of the claims, modifications to the described embodiments are possible, and other embodiments are also possible.

Claims

1. A signal equalizer, comprising: Feedforward filters, which include: Output; The first filter stage includes: A first filter configured to receive in-phase signal components and output filtered in-phase signals; A second filter configured to receive quadrature phase signal components and output a filtered quadrature phase signal; and The second filter stage includes: A third filter is coupled to the second filter to receive the filtered quadrature phase signal; A fourth filter coupled to the first filter to receive the filtered in-phase signal; A first adder circuit, coupled to the first and third filters, is configured to combine the filtered in-phase signal with the output of the third filter to provide an equalized in-phase signal; and A second adder circuit, coupled to the second and fourth filters, is configured to combine the filtered quadrature phase signal with the output of the fourth filter to provide an equalized quadrature phase signal. A decision device having inputs and outputs, wherein the output of the decision device is coupled to a feedback filter; The feedback filter having an input and an output, the input of the feedback filter being coupled to the output of the decision device; and A subtractor having an output and a first input and a second input, the first input of the subtractor being coupled to the output of the feedforward filter, the second input of the subtractor being coupled to the output of the feedback filter, and the output of the subtractor being coupled to the input of the decision device.

2. The signal equalizer according to claim 1, further comprising: A first adaptation circuit coupled to the first filter stage, the first adaptation circuit being configured to change the coefficients of the first filter stage; and A second adaptation circuit is coupled to the second filter stage, and the second adaptation circuit is configured to change the coefficients of the second filter stage.

3. The signal equalizer of claim 1 further includes an adaptation circuit coupled to the first filter stage and the second filter stage, the adaptation circuit being configured to change the coefficients of the first filter stage and the second filter stage.

4. The signal equalizer according to claim 1, wherein the feedback filter is a two-stage composite filter comprising a first stage and a second stage coupled in series with the first stage.

5. The signal equalizer according to claim 1, wherein the feedback filter is a single-stage composite filter.

6. The signal equalizer according to claim 1, wherein the first filter stage has n taps, the second filter stage has m taps, and the feedback filter has t taps, where t is less than n.

7. The signal equalizer according to claim 1, further comprising: A first adaptation circuit is coupled to the first filter stage, and the first adaptation circuit is configured to change the coefficients of the first filter stage; and A second adaptation circuit coupled to the second filter stage, the second adaptation circuit being configured to change the coefficients of the second filter stage, wherein one of the first adaptation circuit or the second adaptation circuit is coupled to the feedback filter and configured to change the coefficients of the feedback filter.

8. The signal equalizer according to claim 1, further comprising: An adaptation circuit is coupled to the first filter stage, the second filter stage, and the feedback filter, the adaptation circuit being configured to change the coefficients of the first filter stage, the second filter stage, and the feedback filter.

9. A signal equalizer, comprising: Feedforward filters, which include: Output; A first filter stage, the first filter stage comprising: A first filter having an input to receive in-phase signal components and an output; and A second filter has an input to receive quadrature phase signal components and an output; and The second filter stage includes: A third filter having an input coupled to the output of the second filter, the third filter having an output; A fourth filter having an input coupled to the output of the first filter, the fourth filter having an output; A first adder circuit has a first input coupled to the output of the first filter and a second input coupled to the output of the third filter; and The second adder circuit has a first input coupled to the output of the second filter and a corresponding second input coupled to the output of the fourth filter; A decision device having inputs and outputs, wherein the output of the decision device is coupled to a feedback filter; The feedback filter having an input and an output, the input of the feedback filter being coupled to the output of the decision device; and A subtractor having an output and a first input and a second input, the first input of the subtractor being coupled to the output of the feedforward filter, the second input of the subtractor being coupled to the output of the feedback filter, and the output of the subtractor being coupled to the input of the decision device.

10. The signal equalizer according to claim 9, further comprising: A first adaptation circuit coupled to the first filter stage, the first adaptation circuit being configured to change the coefficients of the first filter stage; and A second adaptation circuit is coupled to the second filter stage, and the second adaptation circuit is configured to change the coefficients of the second filter stage.

11. The signal equalizer of claim 9, further comprising an adaptation circuit coupled to the first filter stage and the second filter stage, the adaptation circuit being configured to change the coefficients of the first filter stage and the second filter stage.

Images