Quantizer and optical transmitter

Abstract

To provide a technique for reducing quantization noise.SOLUTION: A quantizer includes a clipper that clips a portion of a sample value sampled at a predetermined rate, which exceeds a quantization range, and a noise shaper that determines a plurality of quantization level candidates on the basis of the clipped sample value, and outputs a value obtained by adding minimum noise that minimizes the noise in a low frequency region among noise generated in each candidate to the sample value before clipping as a quantized value.SELECTED DRAWING: Figure 8

Description

Technical Field

[0001] The present disclosure relates to a quantizer and an optical transmitter.

Background Art

[0002] In recent years, in order to achieve flexible data data rate, multi-level modulation methods using PCS (Probabilistic Constellation Shaping) and ultra-high-level modulation for high-capacity short-distance communication are being put into practical use. As the amount of information to be transmitted, especially the baud rate, increases, the requirement for the bit resolution of the DAC (Digital-to-Analog Converter) in the optical transmitter also increases, but there is a limit to the bit resolution of the DAC. Rather, a lower bit resolution of the DAC can suppress power consumption and simplify the design.

[0003] In digital coherent optical communication that combines digital signal processing and coherent optical transmission, quantization noise occurs in the digitization process. Quantization noise is an error in digital conversion. Even when the level and baud rate of the modulation method increase, it is desirable to reduce quantization noise and support a low-bit-resolution DAC.

[0004] FIG. 1 shows a general quantization model 1. The sample value X(n) sampled by the signal processing circuit is rounded up or down to the nearest quantization level within the quantization range. The distance Δ between quantization levels is called the quantization step size. The error between the sample value X(n) before quantization and the value q(n) after quantization becomes quantization noise q noise and has the characteristics of white noise that is flat with respect to the frequency axis.

[0005] FIG. 2 shows a quantization model 2 (see, for example, Non-Patent Document 2). By soft quantization (SQ), three candidates c including the quantization level 2 closest to the sampling output x(n) and the quantization levels 1 and 3 above and below iti-1 , c i , c i+1 is determined. By path metric calculation, the quantization error is calculated for each of the three candidates. By hard quantization (HQ), the candidate with the smallest mean squared error over N sampling outputs is selected, and the quantization value x Q (n) is output.

Prior Art Documents

Non-Patent Documents

[0006]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0007] If quantization noise can be suppressed, even when the modulation level and baud rate are high, an analog voltage signal with less error can be generated by a DAC with low bit resolution. In one aspect, the present invention aims to provide a technique for reducing quantization noise.

Means for Solving the Problems

[0008] In one embodiment, the quantizer a clipper that clips a portion of the sampled value exceeding the quantization range of the sampled values sampled at a predetermined rate, a noise shaper that determines a plurality of candidates for quantization levels based on the clipped sampled values, and outputs, as a quantization value, a value obtained by adding the minimum noise, which is the minimum noise in the low-frequency region among the noises generated in each candidate, to the sampled value before clipping has

Advantages of the Invention

[0009] A technique for reducing quantization noise is realized.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Embodiments for Carrying Out the Invention

[0011] Hereinafter, the configuration and method of the embodiment will be described with reference to the drawings. In the following description, the same components may be denoted by the same reference numerals, and duplicate descriptions may be omitted.

[0012] FIG. 3 is a block diagram of an optical transmitter 10 to which a quantizer 100 of an embodiment is applied. The optical transmitter 10 includes a signal processor 12, a quantizer 100, a DAC 14, a light source (denoted as “LD” in the figure) 16, and an optical modulator 17. The quantizer 100 may be included in an integrated circuit board of a DSP (Digital Signal Processor) 15 together with the signal processor 12, or may be realized by an ASIC (Application Specific Integrated Circuit) different from the signal processor 12. In FIG. 3, the DSP 15 is denoted as “Tx DSP” in the sense of performing transmission-side processing.

[0013] The sample value X(n) sampled by the signal processor 12 is quantized by the quantizer 100 and input to the DAC 14 as digital data. The sample value X(n) is a numerical value representing the amplitude at the sampling point, whereas the digital data output from the quantizer 100 represents integer values such as “0” and “1”. The DAC 14 converts the input digital data into an analog voltage signal. Based on the output of the DAC 14, a drive signal is generated by a modulator driver amplifier (not shown) and input to the signal electrode of the optical modulator 17.

[0014] The optical modulator 17 is, for example, a modulator using the DP-QPSK (Dual Polarization Quadrature Phase Shift Keying) method, and has an IQ modulator formed by a Mach-Zehnder (MZ) interferometer for each of the X polarization and the Y polarization. Four analog voltage signals are output from the DAC 14, and drive signals are input to the I arm and the Q arm for the X polarization, and the I arm and the Q arm for the Y polarization.

[0015] The carrier wave output from the light source 16 is branched into an X-polarization use and a Y-polarization use, and undergoes a phase modulation corresponding to a refractive index change in each arm of the corresponding IQ modulator. An intensity-modulated optical signal is output from the optical modulator 17 by interference by the child MZ interferometer forming each arm and the parent MZ interferometer in which the child MZ interferometers are nested.

[0016] FIG. 4 shows the basic configuration of the quantizer 100 and the signal processor 12. Transmission data 121 is input to the signal processor 12. The transmission data is, for example, a binary bit sequence. The transmission data 121 undergoes error correction encoding processing by the FEC (Forward Error Correction) pre-coder 122 and is mapped to symbol points on a constellation (IQ complex plane) by the constellation mapper 123. At this time, probability distribution shaping of the symbol points may be performed.

[0017] The assistant signal inserter 124 inserts a known assistant signal into each of the symbol points. The assistant signal is used for estimation of the transmission line status such as the OSNR. Pre-emphasis for compensating for signal attenuation in advance and pulse shaping are performed for each polarization by the pre-emphasis and pulse shaper 125.

[0018] The resampler 126 samples the shaped waveform at a predetermined sampling rate. The resampler 126 oversamples at a sampling rate higher than the frequency of the original signal and outputs sample values X(n) from the signal processor 12. Here, n is the number on the time axis. For example, sample values X(n) obtained by oversampling a 60-gigabo pulse train by a factor of 2 are input to the quantizer 100.

[0019] The quantizer 100 includes a clipper 101 and a noise shaper 103. The clipper 101 clips the portion of the sample values X(n) sampled at a predetermined rate that exceeds the quantization range. The noise shaper 103 determines a plurality of candidates for quantization levels based on the sample values clipped by the clipper 101, and outputs, as the quantization value, a value obtained by adding the minimum noise in the low-frequency region among the noises generated for each candidate to the sample value before clipping.

[0020] When a transmission signal exceeding the maximum input voltage range of the DAC 14 is generated, it may be necessary to clip the transmission signal at the quantization stage. However, since distortion occurs in the output of the DAC 14 due to clipping, generally, the sample values X(n) are set to be within the maximum input voltage range of the DAC 14.

[0021] In the embodiment, the sample values X(n) are stretched in the amplitude direction so that the sample values X(n) input to the quantizer 100 are larger than the maximum input voltage range of the DAC 14, that is, the quantization range of the quantizer 100. The stretching of the sample values X(n) may be performed before being input to the quantizer 100, or may be performed inside the quantizer 100. In the latter case, an expander for stretching the input sample waveform in the amplitude direction may be provided inside the clipper 101.

[0022] Clipper 101 clips the portion of the input sample value X(n) that exceeds the maximum input voltage range of DAC14, i.e., the quantization range. The clipping coefficient is set to an appropriate value such as 0.7 (30% clipping), 0.8 (20% clipping), etc. Let the data after clipping be X'(n). Clipping generates an error from the actual sample value X(n), i.e., clipping noise.

[0023] Noise shaper 103 shapes the noise of the data X'(n) after clipping using the sample value X(n) before clipping. The noise of the data after clipping includes clipping noise and quantization noise. Noise shaper 103 extracts candidates for a plurality of quantization levels for the data X'(n) after clipping of the current symbol. The number of candidates is set in advance considering the bit resolution of DAC14. Calculate the difference between the quantization level of each candidate and the sample value X(n) before clipping. This difference is the total noise obtained by adding clipping noise and quantization noise.

[0024] Noise shaper 103 removes the high-frequency components from the total noise through a filtering process that extracts low-frequency components, and selects the quantization level at which the total noise is minimized in the low-frequency region. Thereby, the quantization noise is reduced so as to be compatible with the bit resolution of DAC14. In clipping, the peak of the waveform of the sample value X(n) is cut off, but within the input voltage range of DAC14, the resolution becomes higher, and as a result, the quantization noise is reduced. The cut-off peak is clipping noise, which is added to the quantization noise.

[0025] The difference between the data X'(n) after clipping and the sample value X(n) before clipping can calculate the total of the clipping noise and the quantization noise. For the data X'(n) after clipping, by selecting candidates for the quantization levels of the extracted DAC using shaping techniques, the low-frequency components of the clipping noise and the quantization noise within the input range of the DAC can be reduced. Although the high-frequency noise components are expanded, since the equalizer of the receiving DSP is an ideal filter, the noise outside the main signal spectrum is cut off. Therefore, only the noise contained in the signal spectrum remains, and only that noise component affects the main signal. Below, the specific operation of the quantizer 100 and a more detailed processing configuration will be described.

[0026] Figure 5 is a flowchart of the quantization process performed by the quantizer 100. First, the sample value X(n) of the current symbol is input (S11). The sample value X(n) is amplitude data resampled by the signal processor 12. Next, the sample value X(n) is clipped to obtain the data X'(n) after clipping (S12). If necessary, the waveform of the sample value X(n) may be stretched in the amplitude direction before clipping.

[0027] Next, using the data X'(n) after clipping, a plurality of quantization level candidates are determined (S13), and the noise S noise of each candidate is calculated (S14). The noise S noise is the total value of the clipping noise c noise and the quantization noise q noise . The clipping noise c noise is represented by the difference between the input sample value X(n) and the data X'(n) after clipping. The quantization noise q noise is represented by the difference between the data X'(n) after clipping and the quantization level q' of each candidate. Therefore, the total noise S noise is represented by X(n) - q'. c noise = X(n) - X'(n) q noise = X'(n) - q' S noise = X(n) - q'

[0028] From among a plurality of quantization level candidates, select the quantization level for which the absolute value of the low-frequency noise LF noise included in the noise S noise is the smallest (S15). Add the noise S noise (i.e., the noise with the smallest noise in the low-frequency region) of the selected quantization level to the sample value X(n) before clipping (S16), and output it as the final quantization result (S17). This output is connected to the input of the DAC14. Determine whether there is the next transmission data (sample value) (S18), and repeat steps S12 to S17 as long as there is transmission data. If there is no input of transmission data (No in S18), end the process.

[0029] FIG. 6 is a schematic diagram of the clipping process in step S12. The sample value X(n) that undergoes the clipping process by the clipper 101 of the quantizer 100 is stretched in the amplitude direction so that its amplitude exceeds the maximum input voltage range of the DAC14. The waveform of the sample value X(n) is expanded in the amplitude direction inside or outside the quantizer 100. Among the sample values X(n), the portion exceeding the maximum input voltage range of the DAC14 is clipped, and the data X'(n) within the maximum input voltage range of the DAC14 is obtained. The error generated by the clipping becomes the clipping noise c noise and is.

[0030] FIG. 7 is a schematic diagram of the quantization level candidate determination in step S13. Quantization level candidates are determined for the data X'(n) after clipping. In this example, the quantization level Q(n) closest to the data X'(n), and the quantization levels Q(n)+Δ and Q(n)-Δ above and below it are used as candidates. Here, Δ is the quantization step size.

[0031] FIG. 8 shows a specific processing configuration of the quantizer 100. Among the sample values X(n) input to the quantizer 100, the portion exceeding a predetermined level is clipped by the clipper 101. The noise shaper 103 determines a plurality of, for example, three candidates for quantization levels based on the clipped data X'(n). The candidate quantization levels q' are respectively q' = Q(n) + Δ q' = Q(n) q' = Q(n) - Δ where Q(n) is the quantization level closest to X'(n), and Δ is the quantization step size.

[0032] The noise shaper 103 inputs each candidate quantization level q' and the sample value X(n) before clipping to the subtractor 105, and calculates the difference as the noise S noise for each candidate. By applying a filter to the noise S noise for each candidate, the low-frequency noise LF noise is extracted, and the selector 104 selects the noise S noise for which the low-frequency noise LF noise (min|LF noise | 2 ) is the smallest. The noise S noise (min|LF noise | 2 ) that minimizes the noise in this low-frequency region is hereinafter referred to as "minimum noise" for convenience.

[0033] In the filtering process, the selected minimum noise S noise (min|LF noise | 2 ) is delayed by one symbol cycle by the delay element 107. The multiplier 108 multiplies the minimum noise S noise (min|LF noise | 2 ) one symbol before by the coefficient h and feeds it back to the current noise S noise . The minimum noise S noise (min|LF noise | 2 ) fed back one symbol before is shaped so that the noise in the low-frequency region is minimized, combining the clipping noise and the quantization noise.

[0034] The minimum noise S before one symbol noise (min|LF noise | 2 ) is added by the adder 106 to the noise S of each current candidate. As a result, the high-frequency components of the noise of each candidate are canceled, and the low-frequency noise LF noise is extracted. The selector 104 selects the minimum noise from the current low-frequency noise LF noise and supplies the selected minimum noise S noise (min|LF noise (min|LF noise | 2 ) to the subsequent adder 109 and repeats the feedback process.

[0035] The feedback circuit formed by the adder 106, the delay element 107, and the multiplier 108 forms a one-tap filter 111. The coefficient h set in the multiplier 108 may be referred to as the "tap coefficient h". The one-tap filter 111 functions as a low-pass filter that cuts high-frequency noise or a feedback equalizer by reflecting the noise shaped one symbol before to the current noise S noise .

[0036] The current noise S noise (min|LF noise | 2 ) selected by the selector 104 is added by the adder 109 to the sample value X(n) before clipping. By this addition, the input sample value X(n) is quantized to the quantization level q'(min|LF noise | 2 ) with the smallest noise in the low-frequency region.

[0037] The quantization level q'(min|LF noise | 2 ) is supplied to the DAC 14 as the output of the quantizer 100. Since a digital signal with the smallest error from the sample value X(n) in the low-frequency region is input to the DAC 14, an analog voltage signal with less error is generated even if the bit resolution of the DAC 14 is not very high.

[0038] Rather than simply quantizing the quantizer 100 to the quantization level closest to the amplitude value of the sample value X(n), among a plurality of quantization level candidates, including clipping noise and quantization noise, the noise S noise selects the quantization level at which the low-frequency component of becomes minimum. Therefore, the quantization accuracy is improved.

[0039] FIG. 9 shows the filter characteristics of the noise shaper 103. (A) in FIG. 9 is the frequency characteristic of the 1-tap filter 111. (B) in FIG. 9 is the shaped noise S noise (min|LF noise | 2 ) spectrum, and (C) is the filter characteristic after 1-tap feedback (equalization) of the shaped noise S noise (min|LF noise | 2 ). The horizontal axis in each figure is the frequency (GHz), and the vertical axis is the gain (dB).

[0040] In FIG. 9(A), the general quantization noise a has the property of white noise that is flat with respect to the frequency axis. When the tap coefficient h is increased to 0.4, 0.6, 0.8, 1.0, the attenuation in the high-frequency region away from the center frequency (0 GHz) increases, and the shape of the low-pass filter becomes steeper.

[0041] In FIG. 9(B), the shaped noise S noise (min|LF noise | 2 ) is small in the low-frequency region and large in the high-frequency region. The larger the tap coefficient h, the steeper the spectrum and the smaller the noise in the low-frequency region.

[0042] (C) in FIG. 9 is the spectrum shape after feeding back the spectrum of (B) to the filter characteristics of (A). The spectrum shape of (C) in FIG. 9 approaches flat with respect to the frequency axis compared to (A) and has the properties of white noise. However, the larger the tap coefficient h, the lower the noise level and the less the influence on the main signal. That is, the SNR is improved. However, the tap coefficient should not simply be made large. As will be described later, the optimal tap coefficient can vary depending on the degree of clipping, the sampling rate of DAC14, etc.

[0043] FIG. 10 shows the setting of five quantization level candidates. FIG. 11 shows the processing configuration of quantizer 100A using the five quantization level candidates. Quantizer 100A has a clipper 101 and a noise shaper 103A. Except for the increase in the number of quantization level candidates, the processing configuration of quantizer 100A is the same as that of quantizer 100 in FIG. 8.

[0044] Among the sample values X(n) input to quantizer 100A, the portion exceeding a predetermined level, for example, the maximum input voltage range of DAC14, is clipped by clipper 101. The sample value X(n) may be stretched in the amplitude direction before being input to clipper 101 or inside clipper 101.

[0045] Based on the data X'(n) after clipping, noise shaper 103A determines the quantization level candidates. The candidate quantization levels q' are respectively q' = Q(n) + 2Δ q' = Q(n) + Δ q' = Q(n) q' = Q(n) - Δ q' = Q(n) - 2Δ are set as.

[0046] The quantization level q' of each candidate and the sample value X(n) before clipping are input to subtractor 105, and the difference is calculated as the noise S of each candidate noise and calculated as. The noise S of each candidate noiseBy applying a one-tap filter 111, low-frequency noise LF noise is extracted, and with a selector 104, the noise S noise with the smallest absolute value of the low-frequency noise LF noise (min|LF noise | 2 ) is selected.

[0047] The selected minimum noise S noise (min|LF noise | 2 ) is delayed by one symbol cycle with a delay unit 107 and added to the current sample value X(n) with an adder 109. By this addition, the quantization level q'(min|LF noise | 2 ) with the minimum noise in the low-frequency region is selected. Even if the bit resolution of the DAC 14 is low, the noise within the spectrum of the quantized signal can be reduced, so an analog voltage signal with less error is generated.

[0048] <Effect verification> FIG. 12 is a schematic diagram of a simulation model for effect verification. Signal processing is performed by a transmission DSP 112 to output a sample value X(n). A DAC quantizer 110 performs the quantization described above, generates an analog voltage signal from the quantized digital value, and drives an optical modulator. Noise is added to the optical signal output from the optical modulator based on the OSNR. On the receiving side, the optical signal is converted into an electrical signal, analog-to-digital conversion is performed by an ADC quantizer 210, and digital signal processing is performed by a receiving DSP 212. The bit resolution of the receiving-side ADC is fixed at 5 bits. In this case, the voltage value is represented in 32 levels.

[0049] FIG. 13 shows the comparison results of the OSNR tolerance under Condition 1. Condition 1 is that the bit resolution of the DAC is 3 bits, the sampling rate of the DAC is 2 sps, the main signal is a Nyquist-shaped 16QAM signal, and the roll-off factor is 0.1. The horizontal axis is OSNR (dB), and the vertical axis is the Q value (dB).

[0050] Line A shows the OSNR tolerance when 30% clipping is performed by the quantizer 100 of the embodiment. Line B shows the OSNR tolerance when 30% clipping is introduced into the quantization model 2 of FIG. 2. Line C shows the OSNR tolerance of the quantization model 2 of FIG. 2.

[0051] Applying 30% clipping to the quantization model 2 improves the Q value. However, even when 30% clipping is introduced and the OSNR is increased, it is below the FEC limit (BER 5.2e -5 ). In contrast, with the quantizer 100 of the embodiment, by minimizing not only the quantization noise but also the noise volume in the low-frequency region of the clipping noise, the FEC limit can be exceeded using a low-bit-resolution DAC.

[0052] FIG. 14 shows the comparison results of the OSNR tolerance under Condition 2. Condition 2 is that the bit resolution of the DAC is 2 bits, the sampling rate of the DAC is 2 sps, the main signal is a Nyquist-shaped 16QAM signal, and the roll-off rate is 0.1. The bit resolution of the DAC is lower than that of Condition 1.

[0053] Line A shows the OSNR tolerance when 30% clipping is performed by the quantizer of the embodiment. Line C shows the OSNR tolerance of the quantization model 2 of FIG. 2. The FEC limit is BER 2.3e -2 . By using the quantizer 100 of the embodiment, the Q value is improved by nearly 3 dB compared with the quantization model 2.

[0054] FIG. 15 shows the comparison results of the OSNR tolerance under Condition 3. Condition 3 is that the bit resolution of the DAC is 3 bits, the sampling rate of the DAC is 2 sps, the main signal is a Nyquist-shaped 64QAM signal, and the roll-off rate is 0.1. A modulation scheme with a higher order is adopted compared with Conditions 1 and 2.

[0055] Line A shows the OSNR tolerance when 30% clipping is performed by the quantizer of the embodiment. Line C shows the OSNR tolerance of the quantization model 2 of FIG. 2. The FEC limit is BER 2.3e-2 It is.

[0056] By minimizing quantization noise and clipping noise in the low-frequency region with the quantizer of the embodiment, even when the modulation order is increased, the FEC limit can be exceeded. In quantization model 2, even if the OSNR is increased, the FEC limit cannot be exceeded.

[0057] From FIGS. 13 to 15, the Q-value improvement effect by the quantizer 100 of the embodiment is confirmed. Also, it can be seen that even when using a DAC with a low bit resolution, the required signal quality can be maintained.

[0058] <Setting of tap coefficient h> Next, the setting of the tap coefficient h will be described. The effect of suppressing quantization noise (including clipping noise) varies depending on the value of the tap coefficient h. As described with reference to FIG. 9, generally, the larger the tap coefficient h, the smaller the noise in the low-frequency region, but there is a limit to the noise suppression effect. The noise suppression limit value depends on the noise spectrum shape and the specifications of the DAC.

[0059] FIG. 16 shows the dependence of the clipping strength of the optimal tap coefficient and the DAC sampling rate. The horizontal axis represents the tap coefficient, and the vertical axis represents the Q-value improvement effect (dB). (A) in FIG. 16 shows the case without clipping, (B) shows 20% clipping, and (C) shows 30% clipping, respectively showing the Q-value improvement effect (dB) when the DAC sampling rate is changed. The Q-value improvement effect is calculated based on condition 2 of FIG. 13 described above.

[0060] When no clipping is performed in (A) of FIG. 16, regardless of the sampling speed of the DAC, by setting the tap coefficient h to 0.8, the noise reduction effect of the 1-tap filter 111 becomes the largest.

[0061] In the case of 20% clipping in Fig. 16(B), when the sampling rate of the DAC is 1.3 sps and 1.8 sps, by setting the tap coefficient h to 0.8, a large noise reduction effect is achieved and the Q value is improved. When the sampling rate of the DAC is 1.5 sps, the tap coefficient h can take any value between 0.4 and 0.9.

[0062] In the case of 30% clipping in Fig. 16(C), when the sampling rate of the DAC is 1.8 sps, by setting the tap coefficient h to 0.8, the maximum Q value improvement effect can be obtained. When the sampling rate of the DAC is 1.5 sps, by setting the tap coefficient h to 0.5, the maximum Q value improvement effect can be obtained. When the sampling rate of the DAC is 1.3 sps, the tap coefficient h can take any value between 0.3 and 0.6.

[0063] When using the quantizer 100 (or 100A) of the embodiment, considering the clipping strength set in the clipper 101 and the sampling rate of the DAC 14, the tap coefficient h of the one - tap filter 111 of the noise shaper 103 can be optimized.

[0064] Fig. 17 is a flowchart of a method for selecting an optimal tap coefficient. Prior to the actual service of the optical transmitter 10, the tap coefficient h set in the one - tap filter 111 of the quantizer 100 is optimized. The value of the tap coefficient h is varied from the minimum value to the maximum value (h = 1.0), and based on the shape of the signal spectrum, the amount of noise contained in the main signal is confirmed, and the value of the tap coefficient at which the noise amount is minimized is selected. This process is performed by the quantizer 100.

[0065] First, the tap coefficient h is set in the one - tap filter 111 of the quantizer 100 (S21). Immediately after the start of the process, as an initial value, the minimum value or the maximum value of the range in which the tap coefficient h is varied may be set. With that tap coefficient h, the shaped noise, that is, the noise S noise (min|LF noise | 2 )(min|LF

[0066] Prepare the spectral shape of the signal (S23), and the shaped noise S noise (min|LF noise | 2 ) multiply the signal spectral shape (S24). Thereby, the components within the signal spectrum, that is, the noise components can be known. Calculate the power of the noise component and store it in the internal or external memory of the quantizer 100 (S25). If there are other values to be set for the tap coefficient h (YES in S26), update the value of the tap coefficient h (S27), and repeat steps S21 to S25. If there are no other coefficient values (YES in S26), select the tap coefficient h with the minimum noise power op and (S28), end the process.

[0067] Thereby, the optimal tap coefficient h can be set for the 1-tap filter 111 of the quantizer 100.

[0068] FIG. 18 is a block diagram of tap coefficient optimization using Nyquist spectrum information. When the main signal is a Nyquist pulse, a Nyquist-shaped spectrum may be used as the spectral shape of S24 in FIG. 17. In this case, the signal processor 12A generates a Nyquist pulse.

[0069] The PRBS (Pseudo-Random Binary Sequence) generator 221 generates a pseudo-random bit sequence. The FEC pre-coder 122 adds a forward error correction code to the bit sequence. The constellation mapper 123 maps the bit sequence with the error correction code added to the symbol points on the constellation. The assistant signal inserter 124 inserts a known assistant signal into each of the symbol points.

[0070] Pre-emphasis and the Nyquist shaper 225 perform pre-emphasis and Nyquist shaping for each polarization. The resampler 226 resamples the Nyquist-shaped pulses to match an AWG (Arbitrary Waveform Generator), that is, a PRBS generator. The resampled sample values X(n) are input to the quantizer 100.

[0071] Prior to the actual service of the optical transmitter 10, the quantizer 100 performs the process of FIG. 17 and optimizes the tap coefficient h set in the 1-tap filter 111. In the quantizer 100, coefficient values are sequentially set for the tap coefficient h, and the noise S noise (min|LF noise | 2 ) is multiplied by the spectral shape of the Nyquist pulse to calculate the power of the noise component. Among all the coefficient values, the coefficient value that minimizes the noise power is selected as the tap coefficient h. By using the optimal tap coefficient h in the 1-tap filter 111, the quantization accuracy is improved.

[0072] FIG. 19 is a block diagram of tap coefficient optimization using transmission line transmission characteristic information. When transmitting a signal at a speed faster than the bandwidth limitation of the transmission line, when shaping the signal spectrum based on the transmission characteristics of the transmission line, the transmission spectrum of the transmission line can be used as the spectral information of S24 in FIG. 17. The signal processor 12B performs THP (Tomlinson-Harashima Precoding) processing. THP is a waveform equalization technique that compensates for inter-symbol interference in the transmission line on the transmission side.

[0073] The FEC pre - coder 122 performs forward error correction processing on the transmission data 121. The constellation mapper 123 maps the error - corrected data to symbol points on the constellation. The assistant signal inserter 124 inserts a known assistant signal into each of the symbol points. The THP circuit 234 compensates for inter - symbol interference of the transmission path in advance. For example, the signal spectrum is shaped based on the transmission characteristics of the transmission path by the data encoding technology of each symbol point.

[0074] The pre - emphasis and pulse shaper 125 performs pre - emphasis and pulse shaping for each polarization. The resampler 226 resamples the pulses that have undergone THP compensation and pulse shaping so as to match the AWG. The resampled sample value X(n) is input to the quantizer 100.

[0075] Prior to the actual service of the optical transmitter 10, the quantizer 100 performs the process of FIG. 17 and optimizes the tap coefficient h set for the 1 - tap filter 111. In the quantizer 100, coefficient values are sequentially set for the tap coefficient h, and the noise S noise (min|LF noise | 2 ), which is shaped to be minimized in low - frequency noise, is multiplied by the transmission characteristics of the transmission path to calculate the power of the noise component. Among all the coefficient values, the coefficient value with the minimum noise power is selected as the tap coefficient h.

[0076] With the configurations of FIGS. 18 and 19, an optimal tap coefficient h corresponding to the Nyquist pulse and the characteristics of the transmission path actually used can be set for the quantizer 100. As a result, the effect of reducing low - frequency noise can be improved, the accuracy of the analog drive signal driving the optical modulator 17 can be increased, and the Q - value of the transmission signal can be improved.

[0077] Although the embodiments have been described based on specific configuration examples above, the present invention is not limited to the configuration examples described above. For example, the number of candidates for quantization levels may be set to any appropriate number of 2 or more according to the bit resolution of the DAC 14. The clipping ratio can be appropriately set to 10%, 20%, etc. in consideration of the maximum input voltage range of the DAC 14.

Explanation of Signs

[0078] 10 Optical transmitter 12, 12A, 12B Signal processor 14 DAC 15 DSP 16 Light source 17 Optical modulator 100, 100A Quantizer 101 Clipper 103 Noise shaper 104 Selector 105 Subtractor 106, 109 Adder 107 Delay element 108 Multiplier 111 1-tap filter (filter)

Claims

1. A clipper that clips the portion of a sample value sampled at a predetermined rate that exceeds the quantization range, A noise shaper that determines a plurality of candidates for quantization levels based on the clipped sample values, and outputs, as a quantization value, a value obtained by adding, to the sample value before clipping, the minimum noise in which the noise in the low-frequency region is minimized among the noises generated for each candidate, A quantizer having the above.

2. The noise includes clipping noise generated by clipping and quantization noise generated by quantization, The quantizer according to claim 1.

3. A filter that extracts the low-frequency component of the noise, The quantizer according to claim 1 or 2, further having the above.

4. The filter An adder that adds the minimum noise one symbol before to the noise of each candidate, A delay device that delays the current minimum noise by one symbol cycle, A multiplier that multiplies the minimum noise delayed by the delay device by a coefficient, Including the above, and the output of the multiplier is connected to the input of the adder, The quantizer according to claim 3.

5. The coefficient is set to a value that minimizes the power of the noise according to the spectral shape of the main signal or the transmission characteristics of the transmission path, The quantizer according to claim 4.

6. A selector that selects the minimum noise from the outputs of the adder, The quantizer according to claim 4 or 5, having the above.

7. A second adder that adds the minimum noise to the sample value before clipping, The quantizer according to any one of claims 1 to 6, having the above.

8. A signal processor that outputs the sample value, The quantizer according to any one of claims 1 to 7, A digital-to-analog converter that digitally converts the quantization value, An optical modulator driven based on the output of the digital-to-analog converter, An optical transmitter having the above.

9. The sample value input to the clipper is stretched in the amplitude direction so as to exceed the maximum input voltage range of the digital-to-analog converter, The optical transmitter according to claim 8.

Images