Wireless communication apparatus with phase noise mitigation

[0001] Phase noise is one of the most critically destructive noises affecting the performance of wireless high speed communication systems. In the general application area, a higher speed communication system requires the use of a higher order QAM (quadrature amplitude modulation) system. We need to use more accurate and more recently measured (i.e., low latency) phase noise estimates for a higher order QAM to mitigate the phase noise effect on the signal. The present invention provides an effective phase noise mitigation approach with good accuracy and very low latency.

[0002] The need to increase the channel data rate in data communication systems without increasing the signal bandwidth drives the development toward more spectrally efficient modulation formats. QAM modulation is widely used in digital wireless communication systems for achieving high data transmission rates over relatively narrow signal bandwidth. Digital radio systems designed using such modulation schemes must balance the effects of phase noise from local oscillators with the demodulator parameters in determining overall performance. In a M-QAM system each symbol transmitted contains k bits, where 2^k=M. In contrast, BPSK (binary phase shift keying) system transmits 1 bit per symbol and QPSK (quadrature phase shift keying) system transmits 2 bits per symbol while requiring the same RF (radio frequency) bandwidth for a given symbol rate.

[0003] The M states in a high order M-QAM constellations are more closely spaced than in BPSK or QPSK constellations and therefore require lower noise relative to the average carrier power to eliminate errors. The RMS (root mean square) phase noise of the local oscillators, after being filtered by demodulator, must be sufficiently low to not cause bit errors. For a QPSK or M-PSK schemes (or systems), a multitude of carrier recovery algorithms exist that provide a high phase noise tolerance. However, those algorithms fail when applied to most of the higher order QAM constellations because these lack equidistant phases. Additionally, it has been shown that most of decision-directed carrier recovery is also not a viable option for higher order QAM constellations due to the inevitable relatively long feedback delay (or latency) in practical systems.

[0004] The present invention provides a phase noise tolerant method with very low latency by combining the decision feedback equalizer (DFE) and the carrier recovery loop effectively.

• [0005] Cheng-I Hwang, David W. Lin, “Joint Low-Complexity Blind Equalization, Carrier Recovery, and Timing Recovery with Application to Cable Modem Transmission”, IECE Trans., Commun., Vol. E82-B, No. 1 Jan. 1999

[0006] The present invention is related to a method and apparatus for mitigating phase noise in data communication system. The present invention provides effective phase noise mitigation with very low latency by effectively combing decision feedback equalizer (DFE) and carrier recovery loop.

[0010] The following callout list of elements can be a useful guide in referencing the elements of the drawings.

• [0011] 100 DD/Input Signal of the DFE u(n)
• [0012] 102 Multiplier/Phase Noise Compensator
• [0013] 103 Phase Noise Compensated Input Signal x(n)
• [0014] 104 FFF/Feed Forward Filter
• [0015] 106 Adder
• [0016] 107 Output Signal of the DFE q(n)
• [0017] 108 Multiplier/Phase Noise Compensator/Second Phase Noise Compensator
• [0018] 109 Phase Noise Compensated Signal/Input Signal Q(n)
• [0019] 110 The DD
• [0020] 112 Adder/Error Signal Calculator
• [0021] 113 Error Signal e(n)
• [0022] 114 Filter Coefficient Adaption Block
• [0023] 115 Step Size μ
• [0024] 116 Filter Coefficient Adaption Block
• [0025] 118-FBF/Feedback Filter
• [0026] 120 Data Signal energy sample |{circumflex over (d)}(n)|²
• [0027] 122 First Low Pass Filter
• [0028] 124 Error Energy Estimate of the Error Signal |e(n)|²
• [0029] 126 Second Low Pass Filter
• [0030] 128 Output Signal of the DD {circumflex over (d)}(n)
• [0031] 212 DD Input Signal Normalizer
• [0032] 214 Phase Noise Calculator
• [0033] 216 Complex Conjugator Block/Complex Conjugator
• [0034] 218 Imaginary Part Selector Block/Imaginary Part Selector
• [0035] 220 Low Pass Filter
• [0036] 222 Adder
• [0037] 223 Symbol Delay Operator/Symbol Delay Block
• [0038] 226 Phase Noise Estimate/Phase Noise Estimate Block/Complex Number Conversion Block

Input Signal u(n): An input signal is a noisy signal that is the input of the phase noise compensator. The output of the phase noise compensator is the phase noise compensated input signal x(n).

[0039] Decision feedback equalizer (DFE): A DFE is a nonlinear equalizer that automatically adapts to time varying properties of the communication channel. It is frequently used in coherent modulation systems to mitigate the channel effect and the inter symbol interference in the received data.
Phase Noise Compensator: a complex multiplier that is used to reduce phase noise by multiplying the negative value of the phase noise estimate with the phase noisy signal.
Phase Noise Compensated Input Signal x(n): The phase noise compensated signal is obtained from the output of phase noise compensator. The phase noise compensated signal which is the input to the FFF in DFE is x(n).
Feedforward Filters (FFF): FFF consists of finite impulse response filter. FFF combines delayed versions of signals to mitigate the channel effect in the received input signal.
Adder: adds input signals together to produce an output signal.
Output Signal of the DFE q(n): a signal that outputs from the DFE. The output signal of the DFE, q(n), is used as the input of the phase noise compensator to generate DD input signal Q(n).
Phase Noise Compensated Signal: A signal after it is compensated for phase noise.
Decision Device (DD) Input Signal Q(n): A phase noise compensated signal from the phase noise compensator. It is an input signal that is received into a decision device.
Error Signal Calculator: an adder that calculates an error signal.
Error Signal e(n): The difference between the DD output and phase noise compensated DFE output.
Filter Coefficient Adaption Block: The function block updates the adaptive filter coefficient iteratively. The function block use the error signal, step size, and input signal of the each filter coefficient.
Step Size μ: a small proportional constant which is used in the Least Mean Squared (LMS) algorithm for adaptive filter coefficient update.
FBF/Feedback Filter: Feedback filter is a part of decision feedback equalizer. It is effective to mitigate inter symbol interference cancellation.
Data Signal Energy Sample: |{circumflex over (d)}(n)|²: Energy of data signal sample.
Low Pass Filter: A filter that allows low frequency to pass, but not high-frequency.
Error Signal Energy Sample |e(n)|²: Energy of the Error Signal sample.
Output Signal of the DD {circumflex over (d)}(n): An output signal from the decision device.
Input Signal Normalizer: Function block which divide the input signal by the energy of the input signal.
Phase Noise Calculator: A Multiplier that calculates phase noise sample.
Complex Conjugator Block: Function block which apply the conjugate function on the input complex signal.
Imaginary Part Selector Block: The function block which takes the imaginary part of the input complex signal.
Symbol Delay Operator: A block that introduces a symbol delay to a signal.
Phase Noise Estimate Block: A block that converts a phase noise estimate into a complex number using complex number conversion function.
Equalization

[0040] Equalization is used to mitigate the channel effect on the signal for reliable communication. The equalizer estimates the communication channel distortion (such as amplitude distortion, phase distortion, fading, and interference, etc.) and mitigates the channel effect by compensating the channel distortion on the received signal.

[0041] In micro-wave or millimeter-wave communications that use high radio carrier frequency, the phase noise is high because the amount of phase noise is proportional to the carrier frequency used. In communication systems such as these, one of the most powerful solutions to fight against phase noise is to combine the equalizer with carrier recovery while exploiting the latency reduction between phase noise estimation and correction.

[0042] We use a fractional or symbol spaced decision feedback equalizer (DFE) to mitigate the channel effect and inter-symbol interference. The DFE consists of two sections in equalizer: one is Feed Forward Filter (FFF) and the other is Feedback Filter (FBF). Both the FFF and FBF consist of a finite impulse response (FIR) filter.

[0043] FIG. 1 is the function block diagram of the DFE which gets the phase noise estimate from carrier recovery loop. The input signal of the DFE, u(n) 100, is the transmitted signal through the noisy communication channel from the transmitter and is the signal received at the receiver. This is a noisy signal. The output signal, q(n) 107, of the DFE is the summation of the output signal of the FFF 104 and the output signal of the FBF 118 in the adder 106. The output signal, {circumflex over (d)}(n) 128, of the DD 110 is used in the decoder.

[0044] The phase noise correction (or compensation) is obtained by multiplying the respective input signal of the DFE, u(n) 100, and the output signal, q(n) 107, of the DFE with the negative value of the phase noise estimate 226, which is obtained from the carrier recovery DPLL (digital phase locked loop) in the multipliers 102 and 108, respectively. The multiplier 102 is the name of the hardware component and its function is a phase noise compensator. The phase noise compensated signal, Q(n) 109, is the input of the DD 110. The DD 110 uses the input signal, Q(n) 109, to find its output from M QAM constellation by finding one of the M signal constellation points which is closest to the input signal, Q(n) 109, and selects the closest signal point to Q(N) 109 as the output signal, {circumflex over (d)}(n) 128, of the DD 110. The error signalm, e(n) 113, is obtained by subtracting the output signal, {circumflex over (d)}(n) 128, from the input signal, Q(n) 109, of the DD 110 in the adder 112, which is the error signal calculator.

[0045] The error signal, e(n) 113, the phase noise compensated input signal, x(n) 103, and the step size, μ 115, are used in the filter coefficient adaption block 114 to update the filter coefficient of the FFF 104. The error signal, e(n) 113, the output signal, {circumflex over (d)}(n) 128, and the step size, μ 115 are used in filter coefficient adaption block 116 to update the filter coefficient of the FBF 118.

[0046] Both the FFF 104 and the FBF 118 are asymmetric complex FIR filters. The least mean square (LMS) algorithm can be used to update the filter coefficients in both the FFF 104 and the FBF 118. The step size (μ) used for the filter coefficient update in the LMS algorithm is programmable.

[0047] The filter coefficient adaption blocks of 114 and 116 performs filter coefficient update process based on the following procedures;

a) Calculate the output signal q(n) 107 of the DFE in the adder 106;

[0048] $q\left(n\right)=\sum _{k=1}^K{f}_k\left(n\right)x_k\left(n-k\right)+\sum _{l=1}^L{b}_l\left(n\right){\hat{d}}_l\left(n-l\right)$
b) Calculate the phase noise compensated signal Q(n) 109;
Q(n)=q(n)e^{−j{circumflex over (θ)}n-Δ}={circumflex over (d)}_ne^{j{circumflex over (θ)}n}
where K and L are filter order of FFF and FBF, respectively, θ_n-Δ is the phase noise estimate obtained from carrier recovery loop, and {circumflex over (θ)}_n, is the residual phase noise after phase noise compensation.
c) Generate the output signal, {circumflex over (d)}(n) 128, of the DD 110 based on Q(n) 109 using the selection process mentioned above.
d) Calculate the error signal e(n) 113 by subtracting the output signal {circumflex over (d)}(n) 128 from the phase noise compensated input signal Q(n) 109;
e(n)=Q(n)−{circumflex over (d)}(n),
e) Update the filter coefficients of the FFF 104 and FBF 118 using the LMS algorithm as follows;
f_k(n+1)=f_k(n)+μ*e(n)*x_k(n), k=1,2, . . . ,K
b_l(n+1)=b_l(n)+μ*e(n)*{circumflex over (d)}_l(n), l=1,2, . . . ,L
for FF filter and FB filter, respectively. Where f_k(n) and f_k(n+1) are present k^th FFF coefficient and next k^th FFF coefficient, respectively. The b_l(n) and b_l(n+1) are present l^th FBF coefficient and next l^th FBF coefficient, respectively. The x_k(n) and {circumflex over (d)}_l(n) are the input of filter coefficient of the k^th FFF and the input of l^th filter coefficient of the FBF, respectively.

[0049] The signal to noise ratio (SNR) estimate is obtained by applying a time averaging first low pass filter (LPF) 122 on the data energy estimate, |{circumflex over (d)}(n)|² 120, of the data signal and by applying a time averaging second low pass filter (LPF) 126 on the error energy estimate, |e(n)|² 124, of the error signal using 1-pole IIR (infinite impulse response) filter as follows;
S²(n)=αS²=(n−1)+(1−α){circumflex over (d)}(n){circumflex over (d)}·(n)=αS²(n−1)+(1−α)|{circumflex over (d)}(n)|²
N²(n)=αN²(n−1)+(1−α)e(n)e*(n)=αN²(n−1)+(1−α)|e(n)|²
SNR=10*log 10(S²(n)/N²(n)),
Where is S²(n) the signal energy, N²(n) is the noise energy, and a is an average parameter of the LPF 126. The value of the α is 0<α<1 and is close to 1.
Carrier Recovery

[0050] FIG. 2 is the function block diagram of the carrier recovery loop (CRL) combined with the part of the DFE that is designed for joint detection, estimation, and compensation of the phase noise. In FIG. 2, the carrier recovery loop consists of the multiplier (phase noise compensator) 108, the DD 110, the adder (error signal calculator) 112, the complex conjugator 216, DD input signal normalizer 212, multiplier (phase noise calculator) 214, imaginary part selector 218, low pass filter 220, the oscillator which consists of the adder 222 and one symbol delay operator 223, and the phase noise estimator 226. The CRL is a 2^nd order type II digital phase locked loop (DPLL) operating at the symbol rate. The CRL tracks the phase error at the input of the DD 110 in the DFE. In FIG. 2, the part of DFE consists of the multiplier (phase noise compensator) 102, the FFF 104, the multiplier (phase noise compensator) 108, the DD 110, the adder (error signal calculator) 112, and the FBF 118.

[0051] The signal processing used to obtain the phase noise compensated signal Q(n) 109, the output signal {circumflex over (d)}(n) 128, and the error signal e(n) 113 from the input signal u(n) 100 in FIG. 2 is the same as those explained above in DFE section.

[0052] The input signal u(n) 100 of the DFE is the signal transmitted through the noisy communication channel from the transmitter and is the received signal at the receiver, which is a noisy signal. The output signal q(n) 107 of the DFE is the summation of the output signal of the FFF 104 and the output signal of the FBF 118 in the adder 106. The output signal {circumflex over (d)}(n) 128 of the DD 110 can be used in the decoder.

[0053] The phase noise correction (or compensation) is obtained by multiplying the respective input signal of the DFE, u(n) 100, and the output signal of the DFE, q(n) 107, with the negative value of the phase noise estimate 226, which is obtained from carrier recovery DPLL (digital phase locked loop) in the multipliers 102 and 108, respectively. The phase noise compensated signal, Q(n) 109, is the input of the DD 110. The DD 110 uses the input signal Q(n) 109 to find its output from M QAM constellation by finding one of the M signal constellation points which is closest to the input signal, Q(n) 109, and selects the closest signal point to Q(N) 109 as the output signal 128, of the DD 110. The error signal e(n) 113 is obtained by subtracting the output signal {circumflex over (d)}(n) 128 from the input signal Q(n) 109 of the DD in the adder 112 (error signal calculator). a3

[0054] The CRL uses the DD input Q(n) 109 and the error signal e(n) 113 to obtain the phase noise estimate. The CRL obtains the phase noise estimates based on the following procedure;

[0055] a) Obtain the input signal, x(n) 103, of the FFF by performing the phase noise correction (or compensation) by multiplying the input signal, u(n) 100, of the DFE with the negative value of the phase noise estimate 226 that is obtained from the carrier recovery DPLL (digital phase locked loop);
x(n)=u(n)e^−jθ
b) Obtain the phase noise compensated input signal of the DD (DD input), Q(n) 109, by performing the phase noise correction (or compensation) by multiplying the output signal, q(n) 107, of the DFE with the negative value of the phase noise estimate 226 which is obtained from carrier recovery DPLL (digital phase locked loop);
Q(n)=q(n)e^{−j{circumflex over (θ)}n-}Δ={circumflex over (d)}(n)e^{j{circumflex over (θ)}n}
where {circumflex over (θ)}_n-Δ is the phase noise estimate obtained from the carrier recovery loop, and {circumflex over (θ)}_n is the residual phase noise after phase noise compensation.
c) Generate the output signal, {circumflex over (d)}(n) 128, of the DD 110 based on the DD input, Q(n) 109, using the selection process mentioned above.
d) Calculate the error signal e(n) 113 (or DD_Noise) by subtracting the DD output signal, {circumflex over (d)}(n) 128, from the DD input signal, Q(n) 109;
e(n)=Q(n)−{circumflex over (d)}(n)
e1) Calculate the output signal, q(n) 107, of the DFE in the adder 106;

[0056] $q\left(n\right)=\sum _{k=1}^K{f}_k\left(n\right)x_k\left(n-k\right)+\sum _{l=1}^L{b}_l\left(n\right){\hat{d}}_l\left(n-l\right)$
where K and L are the filter order of FFF and FBF, respectively.
e2) Calculate the normalized DD input, N_Q, by dividing the DD input signal, Q(n) 109, by the energy of the DD input signal, |Q(n)|², in the DD input signal normalizer 212;
N_QQ_n/|Q_n|²
f) Calculate the complex conjugate value of the DD error, CC_DE, by applying the complex conjugate function on the DD error signal, e(n) 113, in the complex conjugator block 216;
CC_DEe·(n)*=(n)−{circumflex over (d)}(n))*
g) Calculate the normalized DD error, N_E, by multiplying the complex conjugate of the DD error signal with the normalized DD input, N_Q, in the multiplier 214;

[0057] $\mathrm{N\_E}\frac{{e\left(n\right)}^{*}Q\left(n\right)}{{Q\left(n\right)}^2}={\left(Q\left(n\right)-\hat{d}\left(n\right)\right)}^{*}\frac{Q\left(n\right)}{{Q\left(n\right)}^2}$
h) Calculate the phase noise sample by taking the imaginary part of the normalized DD error signal in the imaginary part selector 218;

[0058] Phase noise sample

[0059] $\mathrm{Im}\left\{{\left(Q\left(n\right)-\hat{d}\left(n\right)\right)}^{*}\frac{Q\left(n\right)}{{Q\left(n\right)}^2}\right\}$
i) Calculate the phase noise estimate by taking the time averaging on the phase noise samples. The time averaging is achieved by passing the phase noise samples to the low pass filter 220 and then to the oscillator which is consisted of the adder 222 and the symbol delay operator 223 in the digital phase locked loop (DPLL);

[0060] Phase noise estimate

[0061] $E\left[\mathrm{Im}\left\{{\left(Q\left(n\right)-\hat{d}\left(n\right)\right)}^{*}\frac{Q\left(n\right)}{{Q\left(n\right)}^2}\right\}\right]=E\left[\mathrm{Im}\left\{1-e^{j{\hat{\theta }}_{}}\right\}\right]$
≈[−{circumflex over (θ)}_n]=−{circumflex over (θ)},when {circumflex over (θ)}_n is small

[0062] j) Convert the phase noise estimate into complex valued phase noise estimate using the phase noise estimator 226 and then perform phase noise correction (or compensation) by multiplying the respective input signal of the DFE, u(n) 100, and the output signal, q(n) 107, of the DFE-with the negative value of the phase noise estimate 226, which is obtained from carrier recovery DPLL (digital phase locked loop) in the multipliers 102 and 108, respectively, and then generate the input signal, x(n) 103, of the FFF and the input signal, Q(n) 109, of the DD;
x(n)=u(n)e^−jθ
Q(n)=q(n)e^−jθ

[0063] The LPF (low pass filter) in the Digital Phase Locked Loop (DPLL) in the above FIG. 2 is used for phase noise averaging purpose. The LPF used in DPLL is a classic filter. The LPF in the DPLL has two signal paths in it; one is a proportional path and the other is an integration path as shown in the following FIG. 3. As can be seen in the FIG. 3, the proportional path gain K_p and integration path gain K_i in the LFP determines the bandwidth of the carrier recovery loop filter. Note that the T in FIG. 3 represents one symbol delay operator.

[0064] The carrier recovery loop supports a range of loop bandwidths designed to maximize phase noise tracking capability and minimize the effect of Additive White Gaussian Noise (AWGN). We need to optimize the loop bandwidth based on expected Signal to Noise Ratio (SNR) and phase noise.