Patent Grant
10700897
Modern optical or wireless receivers and transmitters may apply optical or analog domain sub-band processing in order to enhance throughput.
There is a growing need to provide efficient optical or wireless receivers and transmitters of high throughput and low complicity.
There may be provided a receiver, a transmitter and a method.
The subject matter regarded as the invention is particularly pointed out and distinctly claimed in the concluding portion of the specification. The invention, however, both as to organization and method of operation, together with objects, features, and advantages thereof may best be understood by reference to the following detailed description when read with the accompanying drawings in which:
It will be appreciated that for simplicity and clarity of illustration, elements shown in the figures have not necessarily been drawn to scale. For example, the dimensions of some of the elements may be exaggerated relative to other elements for clarity. Further, where considered appropriate, reference numerals may be repeated among the figures to indicate corresponding or analogous elements.
In the following detailed description, numerous specific details are set forth in order to provide a thorough understanding of the invention. However, it will be understood by those skilled in the art that the present invention may be practiced without these specific details. In other instances, well-known methods, procedures, and components have not been described in detail so as not to obscure the present invention.
The subject matter regarded as the invention is particularly pointed out and distinctly claimed in the concluding portion of the specification. The invention, however, both as to organization and method of operation, together with objects, features, and advantages thereof, may best be understood by reference to the following detailed description when read with the accompanying drawings.
It will be appreciated that for simplicity and clarity of illustration, elements shown in the figures have not necessarily been drawn to scale. For example, the dimensions of some of the elements may be exaggerated relative to other elements for clarity. Further, where considered appropriate, reference numerals may be repeated among the figures to indicate corresponding or analogous elements.
Because the illustrated embodiments of the present invention may for the most part, be implemented using electronic components and circuits known to those skilled in the art, details will not be explained in any greater extent than that considered necessary as illustrated above, for the understanding and appreciation of the underlying concepts of the present invention and in order not to obfuscate or distract from the teachings of the present invention.
The following abbreviations and reference numbers are used in the following text and drawings:
ADC=Analog to Digital Converter
DAC Digital to Analog Converter
CD Chromatic Dispersion
FDE Frequency-Domain-Equalizer
EQZ Equalizer
OS OverSampled
FD Frequency Domain
SB Sub-Band
rOS r-fold oversampled
OS Over sampled
LPF Low Pass Filter
IR Impulse Response
CR Carrier Recovery
LMS Least Mean Squared
PDM POL Division Muxing
POL Polarization
MSB Multi Sub-band
DFT-S DFT Spread
Tx Transmitter
Rx Receiver
SC Single Carrier
NS-SC Nyquist Shaped Single Carrier
MSBE Multi Sub-Band Equaliz(ed/ization)
SSC Sub Single-Carrier
IQ In-phase and Quadrature components
IQI IQ Imbalance
CFO Carrier Frequency Offset
CTO Coarse Timing Offset
E&C Estimation and Compensation
FB Filter Bank
PDM Polarization Division Multiplex(ed/ing)
LO Local Oscillator
FTO Fine Timing Offset
SFO Sampling Freq. Offset
CR Carrier Recovery
MSDD Multi.Symbol.Delay.Det
COMP Compensator
DEMOD Demodulator
PN Phase Noise
PROC Processor
COH Coherent
FE Front-end
BE Back-end
CP Cyclic Prefix
SC-FDM Single-Carrier Frequency Division Multiplexing
ISI Inter Symbol Interference
ICI Inter Channel Interference
pnt Points (e.g. in size of (I)FFTs)
There is provided a method and system for breaking the digital processing into multiple parallel virtual sub-channels, occupying disjoint spectral sub-bands. The optical community is well-used to this concept in the optical or analog sub-carrier domains, but it turns out that it can also be done efficiently in the digital domain. E.g., in our embodiment, in the optical Receiver (Rx) ASIC DSP, each exemplary 25 GHz WDM channel would be digitally partitioned into M=16 bands of ˜1.6 GHz each. Remarkably, this digital demultiplexing into sub-bands may be performed efficiently, with low computational complexity, with neither spectral guard-bands in-between the 1.6 GHz sub-channels nor with spectral guard-bands between the 25 GHz WDM channels.
We emphasize that the sub-banding scheme is not analog, but it is rather purely digital, performed after A/D conversion for each individual channel in the WDM multiplex, amounting to a second tier of fine frequency division de-multiplexing. Digital sub-banding provides benefits similar to those obtained in optically generating relatively narrowband bands (3-6 GHz) within a super-channel structure, an approach recently increasingly adopted in super-hero OFDM super-channel experiments [X. Liu and S. Chandrasekhar, SPPCOM'11 (2011).] However our digital sub-channel mux requires much simpler, lower cost and energy-efficient hardware. As our sub-banding realization is digital rather than analog, we do away with the cumbersome finely spaced multi-tone generator, and we eliminate a large number of DACs, ADCs, modulators, optical filters, analog optical receivers, which would be customarily used in transmission of a finely spaced (3-6 GHz) analog-generated super-channel. Nevertheless, we enjoy the full benefits of having narrowband frequency-flat sub-bands, which are now digitally (de-)muxed. Furthermore, despite efficiently crowding the multiple 1.6 GHz sub-bands with zero spectral guard-bands, we are nevertheless able to maintain a nearly perfect degree of orthogonality between the individual sub-bands (i.e., eliminate inter-sub-band crosstalk), which would have been impossible in a fine-muxed analog/optical generated super-channel.
The means to perform the digital sub-banding in the Tx and especially in the Rx is the usage of digital signal processing (DSP) structures in the real-time hardware processors, called Filter Banks. In particular, this invention addresses the so-called Uniform Filter Banks, where ‘uniform’ here indicates that the sub-bands of the filter-bank are all of the same spectral shape, and are uniformly distributed along the frequency axis, i.e., are regularly spaced. In the sequel we drop the ‘uniform’ qualifier, referring to uniform FB whenever we say FB. Moreover, this invention refers to oversampled (OS) filter banks. The OS qualifier describes FB configurations whereby the sampling rate of each of the sub-bands is higher than the Nyquist rate (difference between the highest and lowest frequency), whereas the Critically Sampled (CS) qualifier (sometimes referred to as maximally decimated/interpolated) describes the case when the sub-bands sampling rate precisely equals the Nyquist rate per sub-band (that is a fraction 1/M of the overall channel bandwidth, where M is the number of sub-bands).
Mathematical notations: Introducing the relevant mathematical conventions used in this application we shall use the zero-padding notation (x[p])ZP[p]:L→M for a finite sequence with support of L points extended by zero-padding to a support of M points. Notice that the discrete variable p over which the zero-padding is performed is explicitly denoted in the ZP[p] superscript, to distinguish from additional indexes, if present.
The IDFT and DFT are defined by
X[β]=βIDFTpM-pnt{x[p]}=Σp=0Mx[p]
x[p]=pDFTβM-pnt{X[β]}=1/MΣp-0MX[β]
where the right-subscript and left-superscript explicitly denote the input and output variables of the transformation, and we introduced the notation
We denote the modulo-D operation by
[x]D≡x mod Dx=└x/D┘D+[x]D,
for which we alternatively adopt either the regular unipolar convention specifying the division remainder to be non-negative, 0≤[x]D≤D or we alternatively adopt bi-polar a convention, −D/2<[x]D≤D/2. Notice that these definitions also work not only for integer D but also for real-valued D.
We define the linear shift (delay) operator Dvd {X(v)}≡X(v−d) as well as the circular shift modulo R operator Cv[s]
Polyphase components of a discrete-time sequence are denoted as follows. The p-th polyphase component modulo P (P integer) of a sequence x[k] is defined as, x[p]
A sparse polyphase modulo Q definition is defined as an Q-fold upsampled (zero-filled) and q-delayed version of the conventional p-th polyphase:
x[q]
where the upsampling is defined as
↑Q{y[k]}={y[0],0,0, . . . ,0,y[1],0,0, . . . ,0,y[3],0,0, . . . 0,y[4],0 . . . }
and D is the unit delay operator (z−1). Sparse polyphases are notationally distinguished from the regular polyphases by the presence of an up-arrow in the superscript [q]↑Q, reminding that not only are the polyphases taken modulo Q, but also Q−1 zeroes are inserted in between their samples and the resulting sequence is also delayed by q time units. The sparse polyphases satisfy the following decomposition for any sequence:
x[k]=Σq=0Q1x[q]
E.g. the sequence X[k]={x[0], x[1], x[2], x[3], x[4], x[5], x[6], x[7] . . . } is the sum of the following sparse polyphases modulo Q=3:
x[0]
x[1]
x[2]
We also introduce double polyphase components modulo P, Q as follows:
x[p]
Finally we may combine regular and sparse polyphases:
x[p]
Filter bank description: Introducing now the relevant terminology for filter banks, a synthesis (analysis) FB is essentially a collection of M BPFs sharing a sharing common output (summation node) or common input (splitter) with the said filters (referred to as bank filters) preceded (followed) by L-fold up (down) samplers. When M=L the FB is referred to as critically sampled (CS) whereas when M>L the FB is referred to as oversampled (OS).
In this text we shall label the BPFs by the index β or i, and we shall alternatively use both unipolar index notation 0<β≤M−1 as well as bipolar index notation −M/2+1<β≤M/2, as most convenient in each circumstance. The prototype filter refers to the particular filter out the BPF collection of filters which is positioned at mid-band, in the center of the overall channel which contains the M sub-bands. In the complex-envelope domain (referred to the center frequency of the channel) this filter is actually not BPF but is rather LPF. The other M−1 bank filters in the digital domain are obtained by uniform circular shifts of the prototype filter, corresponding to modulations by harmonic tones in the time domain.
In the bi-polar index notation the LPF prototype filter at mid-band is assigned index β=0, whereas the BPFs in the left/right half-band of the channel are assigned positive/negative β index. For even M, which is of interest here, there is the β=0 center index,
negative indexes and
positive indexes out of which
are the mirror images of each other, namely
whereas the last index
although positive, will be seen to correspond to two half sub-bands one which is at the extreme positive frequency and while the other one is at the extreme negative frequency. In this text we shall label the bank filters by both unipolar 0<β≤M−1 and bipolar −M/2+1<β≤M/2 index notation as most convenient in each circumstance.
Let gβ[k] (hβ[k]) denote the impulse response of the βth filter at the Tx (Rx) and let Gβ(z) (Hp(z)) denote the corresponding z-domain transfer functions. In the bi-polar index notation the bank filter . at the index β=0 is baseband, i.e. it is an LPF rather than a BPF. This filter coincides with the prototype filter and its impulse response is denoted g0[k] (h0[k]) for the synthesis (analysis) FBs.
As the FBs considered here are always uniform, the frequency responses of all BPFs have the same spectral shape as that of the prototype LPF, and may be obtained by uniformly spaced spectral shifts of the prototype LPF, formally described in terms of the circular shift operator defined above as Gβ(ej2πv/R)=Cv[βR/M1]
The frequency shift corresponds in the the time domain to the modulation relation
A similar relation holds for the analysis bank filters, with G replaced by H. All BPFs are obtained from the LPF prototype by frequency shifts as follows:
Hβ(ej2πv/R)=Cv[βR/M]
Thus, the βth BPF TF is obtained from the prototype TF by right circular shifting in the frequency domain by, applying a shift equal to βS=βR/M where S=R/M is the spectral extent of each sub-band.
In particular the largest positive shift corresponding to the largest filter index βmax=M/2, is
yielding Gβ
There is provided a system-level applications at the Rx rather than Tx, namely the application OS FBs for high-speed optical transmission, but the disclosed techniques are also applicable to wireless communication. These novel system level embodiments, especially at the Rx side, build upon our additional disclosure of inventive embodiments at the subsystem level for both the analysis and synthesis OS FBs. Our OS FB embodiments (both analysis and synthesis FBs) have lower real-time computation complexity than the structures mentioned above prior art devices for the OS FBs. In particular the OS FBs disclosed in [fred harris] are based on extensive buffer data shifts, having to shuffle data in real-time at the input/output of the polyphase filter arrays with the data shifts in the buffers performed at the system clock rate, which is very costly in real-time ASIC realizations, especially at the high-speed required of optical communication. We first describe multiple alternative novel embodiments of the OS FB sub-system structure, disclosing both analysis and synthesis FBs, followed by system-level applications to realize novel optical receivers and also disclose novel optical Tx-s for single-carrier transmission which provide spectrally efficient Nyquist shaping and can be used with the novel optical Rx-s disclosed here, all of which are based on the OS FBs.
In contrast, in
In addition, it is necessary to combine the interpolation/decimation with digital up-down frequency shift, as shown by the multipliers (modulators) with the linear phase factor
Back to the CS FB structure of
Now, for the OS FB structure of
Measured in terms of the sub-band bandwidth, S, the shift is an integer number of units, β, however measured in terms of the over-sampled band, VS, the shift is by a rational number β/V.
The frequency shift operator corresponding to freq. up-conversion, is formally described as Cv[βS]
In words, we evaluate [β]v, and this yields the number of sub-band spectral units by which the baseband spectrum is shifted to the right at the output of the first block in the Tx, i.e. at the βth input port of the V-fold OS synthesis FB.
We shall exemplify the spectral analysis for three cases of interest, namely V=2, V=4 and V=4/3, the spectral analyses for which are shown in
For V=2 the spectral shift at the V-fold OS synthesis input is [[β]2S]2S, i.e. for even β([β]2=0) we have [[β]2S]2S=[0·S]2S=0 i.e. there is no shift, whereas for odd β we have [[β]2S]2S=[S]2S, i.e. the shift is by one SB unit, i.e. half the 2S oversampled band, turning the spectrum from low-pass to high-pass, as shown in
For V=4 (
For V=4/3 (
{0, −(⅓), −(⅔), ⅓, 0, −(⅓), −(⅔), ⅓, 0, −(⅓), −(⅔), ⅓, 0, −(⅓), −(⅔), ⅓}
Thus, for β=0, 1, 2, 3 the shifts are respectively [0]4/3S, [−⅓S]4/3S, [−⅔S]4/3S, [⅓S]4/3S, and then the cycle repeats.
Analytically considering a general V-value, still following the βth branch, this sparse oversampled band of extent VS enters the ↑L upconverter and is spectrally replicated L times. The L identical spectral images form an extended period of LVS=MS=R, occupying the full channel bandwidth.
At this point there comes the BPF, carving out one spectral image out of the L images. The βth BPF has its passband right over the βth SB, with center at position βS from the spectral origin.
The i-th oversampled band image has its center frequency at position iVS. The index i=iβ of the spectral image capturing the center of the βth sub-band is determined from the condition that −½VS≤βS−iVS<½VS. Writing βS=iβ·VS+[βS]VS where
or equivalently [βS]VS=βS−iβ·VS and recalling from the bi-polar modulo definition that we must have −½VS≤[βS]VS<½VS, it follows that we do satisfy −½VS≤βS−iβ·VS<½VS, i.e. the particular index
is the one sought for the spectral image which captures the center of the βth sub-band. Moreover, we have [βS]VS=βS−iβ·VS, meaning that the shift between the center of the βth sub-band and the center of the iβ image capturing the sub-band center is precisely [βS]VS=[[β]V S]VS. However, this is precisely the spectral shift which was applied at baseband by the up-converting modulator. Therefore, as the ↑L op rigidly shifted the basedband oversampled band to the iβth spectral image, then the spectral shift which was applied at baseband precisely brings the baseband sub-band right into the passband of the βth BPF. Thus each of the up-shifted sub-bands is carved with nearly no distortion (ideally without distortion if the filter passband is perfectly flat), and all the shifted sub-bands are superposed by the adder terminating the V-fold OS synthesis FB. A multiplexed spectrum of M bands, occupying a spectral extent ±MS/2=±R/2 has been generated by the Tx.
Conversely, in the Rx (
The multiplexed spectrum, juxtaposing the M sub-bands in frequency, is presented in parallel to the filter bank of M BPFs, each of which carves a corresponding sub-band, with high precision over the extent of the sub-band (as each BPF is flat over a sub-band interval) however as the filters are allowed to mildly slope down in the two neighboring sub-bands around the pass-band sub-band, then each BPF also picks up out-of-sub-band spectral interference, which nevertheless is well separated from the targeted sub-band. The spectral support of each BPF is designed to be precisely ±VS/2 around its center frequency, such that the (one-sided) pass-band frequency is S/2, the stop-band is VS/2, and the transition band occurs over the interval VS/2−S/2=(V−1)S/2. The spectral rolloff factor, α≡stopband/passband=V, i.e., it equals the oversampling factor. The larger the oversampling factor the milder the filter and less coefficients it requires. However larger oversampling factor means the filter must operate faster, therefore the computational complexity per unit time of each BPF turns out to be invariant in the OS factor, V.
The spectral support delineated by +/− the stop-band frequency of the BPF around the BPF center frequency then occupies an oversampled band of extent VS. This band is manipulated by the ↓L down-sampling, in essence replicated L times and the images are superposed. These images are shifted at regular intervals equal to B/L=MS/L=VS, which are seen to equal to the bandlimitation VS of the identical images to (traced to the original spectrum having been bandlimited by the BPF to VS two-sided). These spectral images then fit together with no aliasing overlap covering the full channel spectral region of L·VS=MS=R. However, consistent with the VS-periodicity of the superposition of shifted images, the sampling rate at the down-sampler output now becomes VS, indicating that we have oversampling by a factor V relative to the output of the M-fold downsampling in the CS FB structure of
The VS-wide spectrum, which was replicated all the way down to around the origin, includes within it (possibly offset from the origin depending on the BPF index, β and the OS factor, V) the clean sub-band with spectral interference around it. It is the role of the down-shifted decimator final stage to extract the clean sub-band and reject the interference. To this end, the spectrum is first shifted to baseband by the frequency down-shifter (multiplication by WM−βLk=WV−βk, then the sharp anti-aliasing filter of the decimator removes the spectral interference around the desired sub-band which now falls at baseband. Finally the decimator replicates and superposes this spectrum V times, filling up the entire oversampled band of extent VS by V images of the sub-band, consistent with the sampling rate being reduced from VS down to S, which is precisely the Nyquist rate (equal to the two-sided bandwidth S or twice the one-sided bandwidth S/2). We notice that the operations occurring in the analysis FB in the Rx are the precise opposites obtained by time reversal, relative to the operations which occurred in the synthesis FB in the Tx. Thus the spectral manipulations of
In detail, ignoring the frequency-shifting initially, in
In the realization of the V-fold decimator of
In
For the modified interpolator at the Tx side (
For the modified decimator at the Rx side (
Actually, these routing fabrics, and their associated circular shifts do not perform actual shuffling of data, which would be cumbersome here we have no buffering and no data transfer from location to location within a buffer when we refer to ‘circular shift’, but there is just mapping of inputs to outputs, achieved by a static permutation of wires connecting the inputs or outputs of the smaller N-pnt (I)FFT to a particular sub-segment of the VN-pnt input or output of the larger VN-pnt (I)FFT.
Notice that if these two routing fabrics (the ZERO-PAD & CIRC SHIFT and the CIRC&SHIFT & SB EXTRACT) were cascaded back-to-back, then they would result in an identity system, as the Rx routing fabric is a time-reversed version of the Tx-routing fabric, i.e., the Rx routing fabric may be obtained by time-reversed traversing the Tx-routing fabric in the opposite sense. Next we describe the particular Tx (synthesis FB) routing fabrics for the special cases V=2,4/3,4. The corresponding routing fabrics for the Rx (analysis FB) are simply time-reversed of versions of the Tx ones.
We mention that in all these embodiments additional half-band shifts exchanging the positive and negative frequencies of the (I)FFTs inputs or outputs could be applied at the Tx, provided their mirror images will also be applied at the Rx in the corresponding (I)FFTs. These half-band shifts will result in different mappings of the routing fiber which are nevertheless equivalent, as the pairs of half-band shifts at the Tx and Rx will cancel each other in the overall cascade. In the specific examples to follow we do not incorporate these additional half-band shifts, nevertheless corresponding versions with different routing fibers may also be readily derived by applying such additional half-band shifts.
For the special case of V=2, the circular shifts at the Tx (
For odd β, as the circular shift becomes N, the two N-pnt FFT output halves are now mapped into the two middle segments of the 2N-pnt IFFT input (which were previously zero-padded in the case of even β), which may be characterized as the high-pass band. Specifically labeling the four quarters of the of the 2N-pnt IFFT input from the top down.by the indexes 0,1,2,3, then in the Tx, for even β, the top N/2-points of the N-pnt FFT output are mapped into quarter-0, whereas the bottom N/2-points of the N-pnt FFT output are mapped into quarter-3. At the Tx, for odd β the top N/2-points of the N-pnt FFT output are mapped into quarter-2, whereas the bottom N/2-points of the N-pnt FFT output are mapped into quarter-1.
Dually, at the Rx (the sub-band processors for the analysis FB), the flow is reversed: for even β, quarter-0 of the 2N-pnt FFT output is mapped into the top N/2-points of the N-pnt IFFT output, quarter-3 of the 2N-pnt FFT output is mapped into the bottom N/2-points of the N-pnt IFFT. For odd β quarter-2 of the 2N-pnt FFT output is mapped into the top N/2-points of the N-pnt IFFT output, quarter-1 of the 2N-pnt FFT output is mapped into the bottom N/2-points of the N-pnt IFFT.
Overall, for V=2 (twice OS FB), the Tx routing fabric may be characterized as ADD into HIGH/LOW HALF-BAND, whereas the dual Rx routing fabric may be characterized as EXTRACT HIGH/LOW HALF-BAND.
For V=4/3 at the Tx (synthesis FB), the circular shifts are by [β]VN=[β]4/3N points. Since {[4β′+0]4/3, [4β′+1]4/3, [4β′+2]4/3, [4β′+3]4/3}={0,1,⅔,⅓} then the right shifts applied by the circular fabric are by the following number of points:
Further incorporating the zero-padding operation, the resulting cascade of zero-mapping and the circular shifts above may be described as follows: Partition the N-pnt FFT output into 6 sixths (each of N/6-pnt) and partition the 4/3 N-pnt IFFT input into 8 eights (each also of N/6-pnt). Then, labeling the 6 sixths from top to bottom as 0,1, . . . , 5 and labeling the 8 eights from top to bottom as 0,1, . . . , 7, the 6 sixths are mapped into the 8 eights, including zero-padding of two of the 8 eights, as described next.
For β mod 4=2 we apply a circular shift of ⅔ N-pnt (i.e. 4 mod 8 eights) to the eights obtained for β mod 4=0. Thus, sixths 0,1,2 are mapped into eights 4,5,6 while sixths 3,4,5 are mapped into eights 1,2,3.
For β mod 4=3 we apply a circular shift of ⅓ N-pnt (i.e. 2 mod 8 eights) to the eights obtained for β mod 4=0. Thus, sixths 0,1,2 are mapped into eights 2,3,4 while sixths 3,4,5 are mapped into eights 7,0,1. This concludes the complete description of the routing fabric at the Tx for V=4/3. As for the Rx side, the mappings are simply reversed with the 8 eights of the 4N/3-FFT output being mapped into the 6 sixths of the N-pnt IFFT input (after dropping two of the eights).
For V=4 at the Tx, circular shifts are by [β]VN=[β]4N points. Then the right shifts applied by the circular fabric are by the following number of points: For β mod 4=0 by 0 points (no circular shifts); For β mod 4=1 by N-pnt;
Further incorporating the zero-padding operation, the resulting cascade of zero-mapping and the circular shifts above may be described as follows: Partition the N-pnt FFT output into two N/2-pnt halves and partition the 4N-pnt IFFT input into 8 eights (each also of N/2-pnt). Then, labeling the two halves from top to bottom as 0,1 and labeling the 8 eights from top to bottom as 0,1, . . . , 7, the 2 halves are mapped into the 8 eights, including zero-padding of 6 of the 8 eights, as described next.
For β mod 4=0 the halves 0,1 are mapped into eights 0,7. For successive values of β mod 4 in increments of unity the output eights 0,7 are shifted by N over the 4N record, i.e. by 2 mod 8 eights at a time.
For β mod 4=1 the halves 0,1 are mapped into eights 2,1. For β mod 4=2 the halves 0,1 are mapped into eights 4,3 . . . . For β mod 4=3 the halves 0,1 are mapped into eights 6,5.
This concludes the complete description of the routing fabric at the Tx for V=4. As for the Rx side, the mappings are simply reversed with the 8 eights of the 4N-FFT output being mapped into the two halves of the N-pnt IFFT input being mapped into the two halves (after dropping 6 of the eights).
Having described the frequency shifted interpolator decimator realizations in
The sub-band OFDM Tx (top of
OFDM transmission, in effect detailing the general OFDM Tx/Rx structure of
Thus, the routing fabric of the sub-band Tx/Rx for V=2 amounts to adding the vector of N OFDM sub-carriers into the high and low half bands of the twice oversampled band of 2N points which is mapped onto the IFFT inputs at the Tx (with an opposite process of extracting half of the 2N band (therefore dropping the complementary half) in order to down-sample by a factor of 2, transitioning from 2N points to N points.
Inspecting
Recall that the FB treated here are of the uniform type, namely all the bandpass filters have identical spectral shape but just differ in their uniform positioning along the band at intervals R/M=S. The commonality between the filter shapes and the fixed spectral shifts from one filter to the next one suggest that it may be possible to share computational resources between these filters. Indeed, this is well known in prior art for CS FBs [,B. Porat, A course in Digital Signal Processing Wiley, 1997] which are amenable to highly efficient implementation using a filter array of M polyphase components of the prototype filter as well as an M-pnt FFT. This implementation is also known in prior art as ‘uniform DFT filter bank’ Here we proceed to extend the efficient implementation from CS FBs to OS FBs.
The processing starts with a S/P (1:L splitting here rather than 1:M in the CS case), then the S/P outputs feed a polyphase filter array but the filters of the array are now arranged as Single-Input Multiple Output (SIMO) filters. Each such SIMO filter may be viewed as a filter bank of {hacek over (M)} SISO filters with common input. The SISO filters forming the p-th SIMO filter are all fed by a common output, which is the p-th output of the S/P. Here the integer {hacek over (M)} is the numerator obtained by reducing the fraction V=M/L, i.e. {hacek over (M)} is given by the ratio of M and the largest common divisor of M and L. Equivalently, we define {hacek over (M)}, Ľ as two relatively prime integers in the same ratio as M and L:V≡M/L={hacek over (M)}/Ľ, e.g. if (M,L)=(16,12), then ({hacek over (M)}, Ľ)=(4,3).
The impulse responses of these SISO filters forming the SIMO filters will be specified below in terms of the polyphase components of the prototype filter impulse response. The SISO filters may be viewed as organized in a 2D array of dimension L×{hacek over (M)}. Each of the L columns of this array of filters comprises {hacek over (M)} SISO filters forming a SIMO filter. We use the index q=0, 1, 2, . . . , {hacek over (M)}−1 as a vertical label for the SISO filters within the p-th SIMO filter. The (p,q)-th SISO filter is denoted h0(p,q)[k], labeled in terms of its coordinates (p,q) in the 2D filter array. According to the teaching of this invention, the (p,q)-th filter is specified as the sparse double polyphase component of the impulse response h0[k] of the prototype filter as follows:
k0(p,q)[k]≡h0[p]
Therefore the SISO filters forming the p0th-th SIMO filter are
{h0[p
Due to the sparse double polyphase nature of the filters, the total number of non-zero taps in this L×{hacek over (M)} array of sparse filters equals the number of taps P, of the LPF prototype filter. All SIMO filters operate at a rate which is a fraction 1/L of the input rate R into the filter bank (equal to the rate at each S/P output), thus the number of taps to be evaluated per second is PR/L and the number of taps to be evaluated per unit input sample is P/L.
The L×{hacek over (M)} outputs of array of filters are organized into {hacek over (M)} rows of L signals each, and each row is presented to one of the {hacek over (M)} ZP&CIRC SHIFT modules, which are vertically stacked above one another in the 3D representation of the figure. The (q0)-th ZP&CIRC SHIFT module operates on the outputs ρ[p]h0(p,q
ZP[q]:L→M}, i.e., it zero-pads the L−pnt vector of outputs to length M i.e. the index p is extended over the range p=0, 1, 2, . . . , M−1 with the subset of values for p=L,L+1, . . . , M−1 set to zero. Then a circular shift with the shift parameter given by [−Lq]M is performed upon the zero-padded M-pnt vector. This completes the description of the ZP&CIRC SHIFT modules.
Next, the signals coming out of the ZP&CIRC SHIFT modules (horizontal rows) are organized into vertical columns and the elements of each column are summed up. The outputs of the M summers are input into an M-pnt IFFT.
We note that each cyclic shifts just represents a fixed permutation of the L-input ports into L of the output ports, meaning that each row of outputs of the polyphase filters array may be just mapped into a permuted way onto M outputs providing the final vector to be applied to the M adders. Thus there is no hardware penalty as data does not need to be shuffled in real-time in actual buffers as in some prior art filter bank implementation. Rather each ZP& CRC SHIFT routing fabric simply represents here a static wiring arrangement of L inputs to M outputs including zero-padding of M−L of the outputs.
The following final mathematical formula compactly specifies our highly efficient implementation of the V=M/L-fold oversampled filter bank:
where ρ[k] is the input of the OS analysis FB, ρβ[k] is the βth sub-band output of the OS analysis FB, ρ[p]
We next consider some specific special cases for various values of the OS parameter, V:
V=M/L=4/3,{hacek over (M)}=4
V=M/L=4,{hacek over (M)}=4
V=M/L=2,{hacek over (M)}=4
Although this could be shown in full generality, in
This figure illustrates the following filters: h0[Lq+p]
hD[Lq+p]
Now the structure is no longer sparse. There are 16 taps (complex multipliers shown by little circles) connected to the successive nodes from left to right along the delay line. Each delay element output (as well as the initial delay element input) feeds a taps. The taps from left to right are given by the following formulas:
h[p]
h0[L+p]
h02L+p[0]=h0[p]
h0[3L+p]
h0[p]
h0[L+p]
h0[2L+p]
h0[3L+p]
h0[p]
h0[L+p]
h0[2L+p]
h0[3L+p]
h0[p]
h0[L+p]
h0[2L+p]
h0[3L+p]
Each of these formulas contains three equal expressions based on polyphases mod-4L, double polyphases mod-4 of polyphases mod-2 and polyphases mod-L.
The drawing considers the specific realization of the four ZP&CIRC SHIFT modules for L=16,M=16, though it is evident how to extend it for any value of M divisible by 4, by partitioning the output ports into four segments of M/4 ports each, and partitioning the input ports into three segments of M/4 ports each. The injective mapping of 3M/4 inputs into M outputs corresponds to the mapping of three segments into four segments in the manner shown.
The processing is partitioned here in terms of polyphases modulo 4 of the P/S output ports and likewise polyphases modulo M/4 of the IFFT input ports. For each value of p modulo M/4 three of the polyphase filters participate, being fed by three consecutive values of the p-th polyphase modulo 4. The four respective outputs of each of the SIMO filters are permuted as shown and are then added up (multiple lines converging on a node denote addition). The four summing junction outputs are mapped onto four consecutive inputs of the IFFT, with indexes separated by M/4 as shown.
Thus, as shown in
Notice that the M/2 non-zero elements in the two M-pnt outputs of the two ZP&CIRC SHIFT modules do not overlap in index. This means that the 2-pnt adders are always presented with a single non-zero input, while the other input is zero, thus the adders may be discarded and the non-zero inputs which were supposed to go into the adders may be directly routed to the input of the IFFTs where the adder was supposed to be connected. This implies that the ZP&CIRC SHIFT modules as well as the adders may be discarded. Instead, as shown in
Even/Odd (sparse) polyphases of the polyphase components modulo M/2 of the prototype filter h0[k]:
h0[p]
h0[p]
This figure may be derived from
As the value of {hacek over (M)}=4 is identical to that obtained for the previously treated V=4/3 case, a routing fabric structure similar to that of
These circular shifts are implemented by the wiring diagrams of the figure, shown for the particular value M=16, though the wiring is generalized in an evident way for any value of M which is a multiple of 4.
In terms of operating rates, starting at rate R, the rate of arrivals of M-pnt blocks is R/M, after the S/P 1:M operation the blocks appear in parallel at the M outputs of the S/P; this rate is maintained throughout the filtering and the IFFT, which produces parallel blocks of M samples at the same rate of R/M per second. This is also the rate at which samples appear in each of the output ports of the IFFTs. Every two ports of corresponding index across the four IFFTs are collected into a parallel pair of signals which are time-division multiplexed by the 2:1 P/S modules, bringing the rate up by a factor of 2 to a 2R/M output rate per output wire (cf. the rate R/M per wire out of an M-pnt CS analysis FB—here the rate is V=2 times faster).
The corresponding elements of four resulting M-pnt vectors at the four modified CS analysis FB outputs are then organized in quadruples (the βth elements of each of the four vectors are paired up together) and the resulting quads of signals are input into 4:1 P/S modules (M of them), such that the βth such module generates the βth FB output ρβ[k].
In terms of operating rates, starting at input rate R, the rate of arrivals of blocks of M points is R/M, after the S/P 1:3M operation the rate of retained blocks is R/(3M), this rate is maintained throughout the filtering and the IFFT, which produces parallel blocks of M samples at the same rate of R/(3M) per second. This is also the rate at which samples appear in each of the output ports of the IFFTs. Every four ports of corresponding index across the four IFFTs are collected into a parallel quad of signals which are time-division multiplexed by the 4:1 P/S modules, bringing the rate up by a factor of 4 to a 4R/(3M) output rate per output wire (cf. the rate R/M per wire out of an M-pnt CS analysis FB—here the rate is V=4/3 times faster).
In terms of operating rates, starting at rate R, the rate of arrivals of M-pnt blocks is R/M, after the S/P 1:M operation the blocks appear in parallel at the M outputs of the S/P; this rate is maintained throughout the filtering and the IFFT, which produces parallel blocks of M samples at the same rate of R/M per second. This is also the rate at which samples appear in each of the output ports of the IFFTs. Every four ports of corresponding index across the four IFFTs are collected into a parallel quad of signals which are time-division multiplexed by the 4:1 P/S modules, bringing the rate up by a factor of 3 to a 4R/M output rate per output wire (cf. the rate R/M per wire out of an M-pnt CS analysis FB—here the rate is V=4 times faster).
While in the OS analysis FB diagrams shown heretofore the signal flow has always been from left-to-right, in these dual synthesis FB figures, the signal flow is reversed, proceeding from right-to-left and applying the duality mapping of elements. Thus, the duality property amounts to time-reversal—reversing the direction of flow through the signal processing structures.
FIGS. A-29,B present the embodiment of a hardware-efficient V=4/3 OS synthesis filter bank. The duality with respect to the hardware-efficient V=4/3 OS analysis filter bank of
rather than the CS rate of
The TDM demuxing means that each of the four inputs out of the 1:4 S/P has ¼ of the rate i.e.
This is the rate at which blocks are collected across the FFT inputs, and also the rate of the filtering array, thus M-pnt blocks appear at each filtering array outputs at this rate. The 3M:1 P/S TD muxing action raises the sampling rate by a factor of 3M, yielding the output rate
The four P/S outputs are superposed at this rate yielding the final output rate R.
This completes the disclosure of oversampled filter bank processing structures. The OS FB modules are next used as building blocks form more complex systems for optical transmission, namely the optically coherent transmitters (Tx) and (Rx), based on OFDM and single-carrier (SC) modulation formats. Overviewing these following embodiments, OFDM Rx-s using oversampled filter banks will be referred to as Multi-Sub-Band OFDM Rx-s. We shall also introduce an OS FB version of the prior-art DFT-spread (DFT-S) OFDM [DFT-Spread OFDM for Fiber Nonlinearity Mitigation Yan Tang, William Shieh, and Brian S. Krongold, IEEE PHOTONICS TECHNOLOGY LETTERS, VOL. 22, pp. 1250-1252, 2010] referred to as MSB DFT-S OFDM. We also introduce Nyquist shaped single carrier Tx and corresponding OS FB Rx embodiments referred to as SC-MSBE as well as a new class of embodiments with variable numbers of sub-single-carriers per channel, using sub-band bonding, referred to as bonded-MSBE.
In this application there is also provided Parallel-to-Parallel P/P module specified by the ratio p:q where p,q are two integers one of which divides the other. The P/P module is specified to take p parallel lines and produces q parallel lines. E.g. when p>q and q divides p, i.e. we ‘down-parallelize’ then you can realize the P/P as the parallel juxtaposition of q P/S modules, each being (q/p):1 sized. Notice that P/P with p:p, i.e. p=q is just a trivial identity system taking p parallel lines as input and presenting the same p parallel lines as output. We shall resort to the p:p P/P module (an identity system) as a graphical means to compactly depict a multiple wires conduit. Many of the block diagrams of this disclosure comprise parallel to serial and serial to parallel converters. These conversion modules may actually be used in their strict defining sense, as conceptual representation of data flow, however in practice the nominal serial inputs or outputs to these P/S and S/P conversion modules are often not serialized at all but are handled in parallel in order to ease the high speed processing. Thus, the p:1 P/S modules are often implemented as p:q P/P modules and the 1:p S/P modules are often implemented as q:p P/P with the q number of parallel ports corresponding to the nominal serial input or output possibly differing from p, selected according to hardware processing convenience. Thus, whenever we mention S/P and P/S in the claims it should be understood that these models may be implemented in practice as P/P.
allowing to accommodate the ADC anti-alias guardband over the 16th sub-band, and leaving 15 sub-bands for processing and detecting the net data modulated over the 25 GHz, i.e., each sub-band has a spectral width of 1/15(25 GHz)=1.66 GHz. The sub-bands with corresponding indexes from the X and Y FBs are paired up to feed identical sub-band Rx-s, in turn detailed in
Adopting a bi-polar indexing convention for the M−1 sub-bands (and not counting the extreme ADC-transition sub-band), the i-th X-sub-band and i-th Y-sub-band outputs of the two FBs are routed to feed the i-th Rx of the sub-band processor array, for i=−7,−6, . . . , −1, +1, . . . , +7. Notice that the i=0 index is skipped, since in our preferred embodiment we dedicate this mid-sub-band to pilot tone transmission, hence this sub-band is not routed to one of the sub-band Rx-s of the sub-band processor array but is routed instead to a mid-sub-band processor, used to extract the transmitted pilot in order to assist in channel estimation and linear and nonlinear phase noise compensation. It is not necessary however to adopt this pilot sub-band strategy, dedicating a whole sub-band to the pilot signal, however in our preferred embodiment we do so.
The sub-band processing associated with the MSB OFDM Rx of
Notice that the uniformly spaced frequency offsets of the successive sub-bands induce, via the CD, a group delay proportional to the center frequency of each sub-band, such that the various sub-bands come out time-misaligned with timing offsets forming an arithmetic sequence. By fitting a straight line to the estimated coarse timing offsets, it is possible to estimate the amount of CD, providing an efficient CD monitoring means, however unlike in a conventional coherent Rx wherein CD monitoring and estimation is a very essential function, here explicit CD estimation is not required at all for Rx operation here (in conventional receivers where where CD monitoring and estimation is indispensable and it is necessary to supply a separate means to estimate CD in the channel in order to correctly set the coefficients of the CD equalizer).
After dropping the CP, which is quite short here—in our exemplary system using N=64 sub-carriers per sub-band (represented by 128 samples after 2×OS by the FB), the CP is just one sample, i.e. 1:128=0.78% overhead), the signal is conditioned for performing the OFDM FFT analysis in order to extract the sub-carrier complex amplitudes. The reason we are able keep the CP very short in our scheme (thus attain high spectral efficiency) is that the CD distortion causing loss of orthogonality of the sub-carriers, need only be considered over the restricted bandwidth of each sub-band. Once the filter bank separates out the sub-bands and processes each one of them separately, each sub-band is an effective narrow-band OFDM system in itself. Thus the delay spread induced by CD is very small per sub-band. This is one key advantage of MSB OFDM which achieves in effect reduced guard interval (reduced CP) without necessitating the heavy FDE pre-FFT channel equalizer required in optical long-haul implementations of OFDM transmission in order to keep the CP relatively short. Such pre-FFT equalizer is eliminated in our MSB OFDM Rx.
Continuing along the processing chain in
Next, as the received X and Y signals are linear combinations of the transmitted X and Y polarizations, 2×2 MIMO equalization is performed per sub-carrier. Each signal pair formed by n-th sub-carrier from the X polarization and n-th sub-carrier from the Y polarization for n=0, 1, . . . , N−1, is passed through a 2×2 MIMO transformation consisting of a butterfly of four complex multipliers as shown. The four complex tap values are set by a polarization adaptive tracking module.
Finally, the POL-cross-talk free sub-carriers for the X and Y POLs, emerging from the array of 2×2 MIMO EQZs, are presented in parallel to two Carrier Recovery modules one for each polarization. In the preferred embodiment we would use Multi-Symbol Delay Detection (MSDD) for the Carrier Recovery method [see our other patent application], however other methods may be used to estimate and mitigate the residual phase noise in each POL of each sub-band prior to decision (decision, i.e. slicing, is performed at the back end of each CR module).
The outputs of the CR+decision modules in the sub-band processor array represent the decided symbols for each of the X and Y polarizations of each of the sub-bands. Demapping each of the QAM sub-carrier symbols and multiplexing all resulting bit-streams yields the total bit-stream representing the equivalent of the 2×25 GBd signal.
Recapping the overall operation and rationale of the sub-band Rx, the processing is significantly improved, in terms of complexity as well as performance, by the fact that each sub-band is narrowband, hence it is frequency-flat. Therefore no memory (no FIR filtering) is required in the MIMO equalization, which may be accomplished by a simple 2×2 matrix consisting of four complex multipliers. Moreover, the sole effect of CD within the spectral extent of each sub-band is essentially a pure delay (corresponding to the group delay of the sub-band at its center frequency), while the distortion due to CD is negligible, provided that the sub-band is sufficiently spectrally narrow (in our exemplary system, the 1.66 GHz bandwidth per sub-band satisfies this condition—as a measure of the negligible CD distortion, the number of taps necessary to equalize CD is less than a single tap for 2000 Km standard fiber over 1.66 GHz spectral window).
Thus, the CTO compensation is simply performed by an integer delay equal to the group delay of the sub-band rounded off to a multiple of the sampling interval. The estimation of CTO is more robust for the narrowband sub-bands than for a conventional OFDM receiver operating broadband over the full channel. Such CTO estimation may be carried out by the Schmidl-Cox algorithm or by the Minn algorithm, both used in wireless communication, as further detailed below in
This completes the description of the MSB OFDM sub-band Rx. The overall Rx processing is seen to consist of a “divide&conquer” approach, with the top layer comprising the two filter banks (one per POL) which perform the “divide” of the wide channel spectrum into multiple narrowband sub-bands. The “conquer” occurs in the sub-band Rx processor array, as the sub-bands are easier to contend with individually, rather than processing the overall full channel at once.
In fiber communication, the CD-induced impairments grow quadratically in bandwidth, thus reducing bandwidth by a factor of M yields a beneficial CD-impairment reduction by a factor of M2. The narrow-band frequency-flat sub-bands also imply that adaptive equalization algorithm are going to better convergence and attain faster convergence rates. Therefore, basing the optical coherent OFDM Rx on the OS analysis filter banks is very beneficial in terms of performance. It may be shown by counting the heaviest processing elements, namely the multipliers, that the OS analysis FB Rx of this figure has a substantial complexity reduction advantage over conventional prior art OFDM Rx.
Novel MSB DFT-S OFDM
Our MSB DFT-S OFDM PDM Rx is driven by a prior art DFT-S PDM OFDM transmitter. Both our MSB OFDM and MSB DFT-S OFDM embodiments use the same top level Rx structure as shown
The NS-SC Tx is innovatively realized here as a special case of DFT-S OFDM in the unconventional limiting case of having just one sub-single-carrier, with a few adjustments in terms of the sub-carriers mapping, comprising an innovative method of pilot sub-band insertion. As shown in
K−=(M−2)N=14N=896,K+=MN=16N=1024,K0=N=64=KENDS,M=16
K−=(M−2)N=16N=1024;K+=MN=18N=1152,K0=N=64=KENDS,M=18
K−=(M−4)·N=16N=1024;K+=MN=20N=1280,K0=2N=128=KENDS,M=20
In the following figures we assume the first exemplary design with 896-pnt DFT-spread FFT feeding into the 1024-pnt FFT. This designs apparently has somewhat lower complexity than the other two designs. The pilot and the end intervals are then precisely one sub-band interval wide. This Tx is compatible with SC-MSBE Rx of
Notice the permuted mapping of output ports of the FFT into the input ports of the IFFT, including the insertion of the pilot, as well as the alternating sign modulation by the (−1)k, all intended to work together such that a time-domain input of a certain frequency at the FFT input is mapped into the same time-domain output (albeit upsampled) at the DAC inputs.
The Rx disclosed in
The top level receiver structure of
The Analysis FB separates out the channel into sub-bands. The demultiplexed sub-bands are processed in the array of MSB processors, having the timing of all sub-bands mutually aligned, and removing other impairments, including POL cross-talk. Then the ‘cleaned-up’ sub-bands are re-assembled to form a cleaned-up overall single-carrier filling the entire channel. This is accomplished by a Synthesis FB, which is the dual of the analysis FB, and may also be efficiently realized using any of our embodiments of novel 2×OS analysis FB. Each sub-band processor (PROC) unit of the sub-band processor array feeds a corresponding output of one of the two synthesis FBs for the X and Y POLs, as shown. The i-th sub-band PROC X output feeds the i-th input port of the X-POL synthesis FB, while the i-th sub-band PROC Y output feeds the i-th input port of the Y-POL synthesis FB. The output of each synthesis filter bank (one per POL) is finally processed in a wideband carrier recovery (CR) system, which generates the decisions. Our preferred embodiment for the CR is also MSDD based, but we use a block-parallelized version of this MSDD called poly-block. One advantage of bringing the signal back to higher rate (by re-assembling the sub-bands by means of the synthesis filter bank) prior to applying the MSDD, is that the enhanced sampling rate implies better tolerance to laser phase noise (laser phase noise tolerance is a monotonically decreasing function of the sampling rate, due to the random walk of the phase which picks up more variance over longer sampling intervals, decorrelating the phase samples and degrading the quality of the phase estimation).
vnX=DFT64{WXX·IDFT64{unX}+WYX·IDFT64{unY}}=WXX·DFT64{IDFT64{unX}}+WYX·DFT64{IDFT64{unY}}=WXX·unX+WYX·unY
vnY=DFT64{WXY·IDFT64{unX}+WYY·IDFT64{unY}}=WXY·DFT64{IDFT64{unX}}+WYY·DFT64{IDFT64{unY}}=WXY·unX+WYY·unY
The significance of this result is that the back to back N-pnt (I)FFTs may be discarded in
The sub-bands re-assembly (or synthesis) K−-pnt IFFT at the top level of the SC-MSBE Rx of
We reiterate that the top level SC-MSBE Rx of
Also notice that the mid-sub-band which was interspersed by our NS-SC Tx with the data the inputs into the K+-pnt FFT is routed to the Mid-sub-band processor, hence no sub-band processor is provided for it in the sub-band PROC array, thus the remaining data-carrying sub-bands become contiguous, indicating that the insertion of the pilot sub-band in the SC Tx of
Next we present a family of embodiments referred to as bonded-MSBE (B-MSBE). This is a variant of our MSB DFT-S OFDM, enhanced such that the number NSSC of DFT-S sub-single-carriers (SSC) need no longer coincide with the number M of sub-bands (in our MSB DFT-S OFDM embodiment each sub-band carried a sub-single-carrier as generated by a prior art DFT-S OFDM Tx with precisely M DFT-S FFTs (same number as the number of sub-bands in the OS analysis FB based Rx). The new B-MSBE Rx may also be viewed as a generalized version of an DFT-S OFDM Rx wherein the processing of the individual SSCs is carried out by assigning multiple sub-bands per SSC (rather than precisely one sub-band per SSC as in our earlier MSB DFT-S OFDM embodiment).
In anticipation of comprehending the reception of this signal by the B-MSBE Rx disclosed in
The Rx disclosed in this figure differs from the MSB DFT-S OFDM Rx (
We now have a continuum of Rx structures from conventional OFDM via DFT-S OFDM (viewed as multiplexed transmission of multiple SSCs) all the way to single-carrier transmission, with number of SSCs selected at will to be any divisor of M from 1 to M, and with an underlying sub-band processing structure underneath, which in turn is constructed over an OFDM sub-carrier infrastructure.
These are key elements in the SC-MSBE sub-band PROC.
Describing now the IQI COMP module, this novel module is specified by the following non-linear transformation of the two sub-band signals, comprising linear combinations but also complex conjugations (CC), denoted by an overbar here (making it non-linear):
Rk(i)=
Rk(−i)=−{circumflex over (γ)}(i)
The parameters {circumflex over (γ)}(±i), representing the amounts of IQI imbalance in the two respective mirror sub-bands, are estimated by the IQI EST module. The justification for this IQI COMP structure is that the model for the generation of IQI imbalance impairment in the coherent optical Rx front-end is as follows. Considering all sub-band spectral slices as being conceptually superposed to form the optically detected broadband signal at the optical Rx FE input, then this signal is written as
In particular, the input sub-bands ρk(=i) generate their own IQI impairments at the output of the Rx FE, yielding the following impaired contributions to the overall broadband output rk of the Rx FE:
ρk(±i)IQI=α(±i)ρk(±i)+β(±i)
The first equality above is a well-known model of IQ imbalance, e.g. [J. Tubbax et al., Proc. GLOBECOM'03, (2003)]. The proportionality relation is obtained by dividing and multiplying by α(±i) and introducing the definition of γ(±i) as a single complex parameter defining the imbalance, as constants α(±i) in the E&C chain do not matter (the carrier recovery and polarization demux modules compensate for any undermined multiplicative constants in the processing chain). Now the signal components ρk(±i)IQI within the broadband signal rk at the Rx FE output are no longer confined to the respective ±i sub-bands. Indeed, considering first ρk(i)IQI ∝ρk(i)+γ(i)
rk(i)=ρk(i)+γ(−i)
This completes our model, derived for the first time to our knowledge, of the interaction between IQI and sub-band filtering via a FB. At this point we may verify that our novel embodiment of IQI COMP as described in Eq. (4) is indeed able to compensate for the IQI impairment of Eq. (6). This is accomplished by substituting Eqs. (6) into Eqs. (4) and simplifying, yielding
Rk(i)=ρk(i)+γ(−i)
Rk(−i)=−{circumflex over (γ)}(i)
It is now apparent that if our estimates of IQI parameters are precise, i.e. if γ(±i)≅{circumflex over (γ)}(±i) then the coefficients (γ(±i)−{circumflex over (γ)}(±i)) multiplying the IQI-cross-talk terms
At this point we consider whether the presence of CFO affects the IQI COMP procedure just described. It turns out that it does not, as can be seen from the following argument: CFO due to the offset between the optical LO frequency and the incoming carrier frequency is a common phase impairment identically affecting all sub-bands, amounting to multiplication by the linear phase factor ejθk. Thus ρk(±i) is to be replaced by ρk(±i)ejθk in the derivation above. Defining effective bandpass signals ρ′k(±i)=ρk(±i)ejθk and redoing the derivation above with ρ′k(±i) replacing ρk(±i) everywhere, it is seen that the derivation still holds. The only caveat is whether the ρ′k(±i), which have their spectra shifted from their nominal positions, make it without distortion through the bank filters. We assume small CFO (as can be achieved by tracking the CFO by feedback to the laser or digital feedback to a demodulator ahead of the FB, essentially using a PLL), and moreover we recall that the roll-off of the filter banks is not sharp but is mildly sloping out-of-sub-band. Therefore, the slightly shifted spectra, as induced my small CFO, will make it through the bank filters, and the (I)FFT based decimation will also allow the slightly shifted spectra through, by virtue of the cyclic prefix guardband. Next, we may consider whether the simultaneous presence of CTO will affect the IQI COMP operation. In this case we may again define new effective signals ρ″k(±i)=ρk τ
This completes the proof that the disclosed IQI comp structure indeed achieves compensation of IQI even in the presence of CFO+CTO, provided that reasonable estimates of the IQI parameters may be obtained. Next we disclose in
The novel disclosed IQI EST training sequence procedure actually also assists in joint estimation of CFO and CTO, but those aspects will be described further below. To overview the novel data-aided algorithm for joint estimation of IQI, CTA and CFO, this algorithm is based on training sequences with half their spectrum nulled out at a time, first the lower half then the upper half, which results in the filter bank diverting the IQI interference to the mirror-image output port, where it may be readily estimated by comparison with the unperturbed signal in the original output port, on which the SCA may be further applied to extract CTO and CFO with high precision. Our joint estimation procedure is based on periodically launching training sequences, transmitted into either the upper or lower sub-band while nulling the other sub-band. A training sequence {tilde under (A)}k(±i)TR is launched into either the +i or −i sub-bands at the Tx but not in both; upon the TS being launched in the ±i sub-band then a null is simultaneously launched in the ∓i minor sub-band. Actually all pairs of signals are treated this way simultaneously, meaning that TS are first launched into all sub-bands with index i>0 then into all sub-bands with index i<0. As a result a broadband upper-single-side-band (USSB) occupying half of the channel spectrum (to the right of the optical carrier) is generated, followed by a broadband lower-single-side-band (LSSB) signal occupying the lower half of the channel spectrum (to the left of the optical carrier). The USSB launch is followed by the LSSB launch or viceversa.
Each training sequence (TS) launched at the Tx into a particular sub-band port (while nulling out its minor sub-band input) is designed to serve for joint estimation of all three impairments, and in particular to assist in estimating CTO according to the Delay&Correlate (D&C) principles (e.g. using the Schmidl-Cox Algorithm (SCA) or Minn CTO estimation algorithms used in wireless communication). The training sequences then consist of pairs of related consecutive or separated training symbols, in particular for SCA we shall consider for the pairs of identical OFDM symbols or two identical halves of a single OFDM symbol, with each of the two identical records having length L, whereas for the Minn algorithm we shall consider four consecutive records with the first two records identical, and the last two records identical to the first two up to a sign inversion.
Assuming no IQI is generated at the Tx (and the optical links also do not generate IQI, and also ignoring the link nonlinearity) then we receive at the Tx input the following pairs of signals, in the spectral ranges corresponding +i and −i sub-bands, for the US and LS launches respectively: [{tilde under (ρ)}k−τ
[{tilde under (ρ)}k−τ
In contrast, when data is transmitted rather than U/L-SSB sequences, the pair of FB mirror outputs is given, per Eq. (6) by
[ρk(i)+γ(−i)
Significantly, under U/L-SSB launch, the +i/−i sub-band receives it signal free of IQI, whereas its mirror sub-band −i/+i just receives interference from the original sub-band conjugated, scaled by the appropriate IQI coefficient. Thus the conjugate leakage due to the IQ imbalance is routed to the mirror-image sub-band, resulting in the original sub-band launched signal and its scaled CC interference being separated out to two different sub-band outputs of the FB. In each of the two training cases, both filter bank mirror outputs experience a single delay and a single CFO (albeit of opposite sign). This is in contrast to the case of concurrent excitation of the two filter banks wherein each filter-bank output would contain a mixture of two delays and a mixture of the two opposite sign CFOs. This suggests simple data-aided joint processing of the two filter-bank outputs obtained under the respective USSB- and LS-launches, jointly estimating IQ imbalance, CTO and CFO. We note that the two +i/−i FB outputs are CC of each other up to a constant. To estimate the IQ imbalance parameters in the data-aided training mode, we simply divide one filter bank output by the CC of the other one, yielding:
This is compactly written as
IQIestim: {circumflex over (γ)}(±i)=rk(∓i)/
Remarkably, the conjugate-divisions cancel out the common delay and common CFO present in the two sub-bands under either launch (irrespective of the value of the delay and CFO), while recovering just the IQ imbalance parameters. As the last equation holds for each of the discrete-time instants, k, over the duration of the US/LS-launch training sequence, and since the noises in the different samples are largely independent, it is possible to average the conjugate ratios over the 2 L samples of the training sequence duration, yielding the following improved estimates for the IQ imbalance parameters in the two sub-bands:
where T is the total duration (in discrete-time units) of the training sequence (e.g., consisting of two symbols for the SCA (T=2 L) and of four symbols for the Minn Algorithm (T=4 L)).
This completes the mathematical description of the IQI EST module, the block diagram of which is shown in
It remains to describe the joint estimation of CFO and CTO, occurring in the ‘D&C based CTO&CFO joint EST’ module, for which two possible alternative embodiments are detailed in
where these formulas are particular forms of the following general moving average autocorrelation definition with a lag of L over a moving window W:
For the Minn algorithm module of
Back to
We now consider the usage of the feedforward phase estimate, ψ0P[k], input at the bottom of the JOINT IQI, CFO, CTO E&C module in order to demodulate CFO and (NL) phase noise. As shown in
between them or even generate weighted moving average weighted combinations, though the simplest strategy is to use one signal or the other. The feedforward common phase signal ψ0P[k] is available right away, and can also correct other types of phase impairments in addition to CFO, as explained in
This completes the description of all modules in
Synchronization from IQI EST independent of the CTO EST
One issue which arises is how the receiver knows when the two types of training sequences (U/L-SSB) are transmitted, as only then does the IQI EST module generate a valid output one of its output ports or another. One could suggest that the output of the CTO module provides proper synchronization of the training sequences, however this seems somewhat of a ‘chicken&egg’ problem, as unless IQI is compensated first then the results of the CTO estimation are not accurate, and viceversa. However, the system may still converge as even with an imperfect CTO estimate the moving average of Eq. (10) will yield a relatively precise estimate of IQI even if the averaging window is somewhat off (provided the window is sufficiently long). Once this estimate is derived, the next TS launch will experience less IQI and therefore the CTO estimate will improve, and so forth, indicating potential convergence of the system despite the coupling between the two impairment types.
Yet another approach to synchronizing the TS window for IQI estimation is to realize that the event of SSB TS transmission may be detected by the FB based system which acts as a coarse spectrum analyzer. Thus, a strong imbalance in the powers detected in minor sub-bands is an indication of the presence of training sequence. In fact we already provide the computational means to detect such imbalance, as we take the conjugate ratio of the signals in mirror sub-bands and average all signals. If we were to monitor the powers (absolute squares) of the averaged ratios corresponding to our estimates {circumflex over (γ)}(±i), i.e. the quantities |{circumflex over (γ)}(±i)|2 or even better their sum, then we would experience a sharp drop in value whenever the moving average window coincides with the TS window. To understand this notice that from Eq. (6) it follows that the ratio rk(−i)/
(Here we do not interpret {circumflex over (γ)}k(i) as IQ imbalance but rather as the metric obtained by taking the indicated ratio).
When SSB signals are generated, this expression reduces to Eqs. (8) as seen above. However, when the data is on and the launch is two sided, then this expression is approximated by dropping the IQI interference terms (as the γ(=i) coefficients are small).
The power of this squared ratio signal, still fluctuates, mainly under the influence of the randomness of the transmitted signal, however the average value of this ratio is unity, and under the moving average, this signal is likely to converge to unity with fairly high accuracy.
Alternatively, we may generate the absolute square of the sample-by-sample ratio, then apply an MA, in effect reversing the order of squaring and averaging:
Thus, when we are perfectly synchronized with the TS then either of these signals will be low whereas when when there is no overlap with the DS then either of these signals will be very close to unity. Thus, as the moving window advances, the system will go through a minimum, which will indicate when the window is synched, then it will come back up again/
By generating the mirror image ratios,
we may obtain the timing under the LSSB launch, which corresponds to the CTO for the −I sub-band. Thus, we have two alternative CTO estimates:
|{circumflex over (τ)}γ(±i)|2≡argk min{|{circumflex over (γ)}k(±i)|2};|{circumflex over (τ)}G(±i)|2≡argkmin{|Gk(±i)|2} (13)
The description above discloses a novel method to derive CTO by an alternative method differing from the D&C based methods (SCA and Minn algorithms), free of the chicken&egg coupling between CTO and IQI.
In additional embodiments (not shown) it is possible to further merge this CTO estimation method with the D&C methods (SCA and Minn algorithm) e.g. by subtracting the moving average metrics of Eq. (11) and (12) from the D&C based metrics and taking the argmax of the difference in order to derive improved coarse timing estimates.
Detailing the Tx-side sparse pilot generation, the mid-sub-band (in our sub-bands bi-polar labeling convention the 0th sub-band) for each of the X-POL and Y-POL paths at the Tx is fed by a pilot signal positioned in frequency either at the center (DC) or somewhat offset from the center of the sub-band. The rest of the band is null. This is achieved for OFDM transmission (
In order to mitigate potential generation of Stimulated Brilloin Scattering (SBS) due to the intense pilot tone, we may optionally apply at the Tx dithering of the laser frequency, typically at a rate of at least several MHz. This would yield the functional form ej(θ
ΣlβdTx[l] sin(θdTx[l]k+φdTx[l])
Considering now the structure for reception and processing of the pilot tone in the Rx, main processing path in the mid-sub-band PROC receives the mid-sub-band signal r0X/Y[k] of each POL (we follow the Y-POL in the drawing, as for the X-POL the processing structure is identical) and applies IQI COMP, correcting for the IQ imbalance of this sub-band. For simplicity the preferred embodiment of the mid-band IQI COMP does not generate its own estimation of the IQI parameter {circumflex over (γ)}(0). Rather this parameter is simply obtained as the arithmetic mean) {circumflex over (γ)}(0)=½({circumflex over (γ)}(1)+{circumflex over (γ)}(−1)) of the two IQI parameter estimates from the two i=±1 neighboring sub-bands. In the optional case of partial sparse pilot (i.e., when the sparse pilot fills up just part of the mid-sub-band) a low-pass or band-pass filter is further inserted to extract perturbed pilot band (this is the pilot and the sparse band around it, as injected at the Tx, perturbed over propagation by noise and impairments) generating a signal denoted by ρ0X/Y[k]. The phase ∠ρ0X/Y[k] of this signal is extracted and passed to the ‘pilot offset+optional dither phase cancellation stage’ which generates and subtracts phase waveform corrections related to the pilot frequency offset (the pilot subcarrier spectral position relative to the sub-carrier at DC which is mapped to the Tx laser carrier) and optional dither phase if present, as specified further below. The resulting phase ψ0P[k] is output as a feedforward common phase estimate to all M−2 sub-band Rx-s/PROCs providing (NL) phase & CFO compensation per sub-band.
In addition, the common phase signal is also applied to the stage named CFO PLLs which provides a pair of parallel phase-locked-loops (PLL) estimating CFO by heavily low-pass filtering the common phase signal ψ0P[k] (which includes the phase ramps due to the CFO terms) feeding back the filtered CFO estimates to the LO laser analog frequency input and to a digital demodulator at the input of the filter bank.
Notice that this method is a feedforward one, requiring precise synchronization of the paths experienced by the i-th sub-band and the zeroth sub-band such that they arrive aligned in time, allowing to cancel the common phases. By applying sufficient appropriate relative delays to the data sub-bands and the mid-sub-band output ports of the analysis FB, this feedforward time-alignment condition may be readily met for all the sub-bands.
We now analyze the phase noise and frequency offset mechanisms affecting the phase of the perturbed pilot, explaining why this mid-sub-band processor, operating in conjunction with the processing activated by it in the rest of the system, substantially mitigates phase noise and CFO. In order to improve SNR, it is desirable to have the amplitude Ap of the sparse pilot tone as high as possible, at least higher than the average amplitude of the data-carrying sub-carriers.
As a result of optical propagation, the sparse pilot acquires optical phase noise and amplitude modulation, due to the transmit and receive (LO) laser phase noises as well as due to the linear and non-linear phase noises induced in the optical link, the received sparse pilot signal is given by ρ0X/Y[k]=A0[k]ejψ
ψ0[k]≡θCFOk+θpk+ΣlβdTx[l] sin(θdTx[l]k+φdTx[l])+ϕ0X/Y-FWM[k]+ϕ0X/Y-SPM[k]+ϕ0X/Y-XPM[k]+ϕ0X/Y-ASE[k]+ϕ0Tx-LPN[k]+ϕ0Rx-LPN[k] (14)
where θpk is the phase ramp due to the frequency offset of the pilot tone frequency relative to the Tx laser frequency; θCFOk is the phase ramp due to the offset of the Tx laser frequency relative to the LO laser frequency (thus the frequency offset between the received pilot and the LO induces a phase increment of θCFO+θp per discrete-time unit); βdTx sin(θdTxk) and βdRx sin(θdRx k) are the phase dithers applied at the Tx (either for SBS suppression or due to etalon dithering in the Tx laser and at the Rx due to etalon dithering in the LO laser); ϕ0X/Y-FWM[k], ϕ0X/Y-SPM[k], ϕ0X/Y-XPM[k] are respectively non-linear phase noises (NL-PN) due to inter-sub-bands Four-Wave-Mixing (FWM), intra-mid-sub-band Self-Phase-Modulation (SPM) and inter-sub-bands Cross-Phase Modulation (XPM) all induced within the zero-th sub-band to the interaction of all other sub-bands of the given channel as well as all co-propagating channels in the WDM multiplex; ϕ0X/Y-ASE[k] is linear phase noise induced due to the ASE additive noise injected by all optical amplifiers along the link; ϕ0Tx-LPN[k], ϕ0Rx-LPN[k] are the laser phase noise induced by the Tx laser and the LO Rx laser. Moreover, the term θpk, associated with the offset of the mid-band pilot tone is evidently missing in this expression.
If the total phase impairment were common to all the sub-bands, then we would have ideal phase noise and frequency offset cancellation. However, in practice, the phase noises affecting the various sub-bands are not fully correlated. Let us represent the phase and frequency offset impairment of the i-th sub-band by conducting a similar analysis for the total phase noise ψiX/Y[k] affecting the i-th sub-band signal ρiX/Y[k]=Ai[k]ejψ
ψi[k]≡θCFOk+ΣlβdTx[l] sin(θdTx[l]k+φdTx[l])+φiX/Y-FWM[k]+φiX/Y-SPM[k]+ϕ0X/Y-XPM[k]+φiX/Y-ASE[k]+ϕ0Tx-LPN[k−τ(i)]+ϕ0Rx-LPN[k] (15)
Notice here that some of the subscripts, but not all, have been modified from 0 to i, indicative of some the noise sources being sub-band dependent; those phase noise sources which are constant across all sub-bands qualify as ‘common phase’ (in the sub-band domain). In particular, the LO laser phase noise ϕ0Rx-LPN[k] is a common phase contribution as it equally affects all sub-bands by the nature of the modulation process, as the time-domain superposition of received sub-bands is mixed with the LO laser field ∝ exp{jϕ0Rx-LPN[k]}, meaning each sub-band is individually mixed by the LO laser field. However, the Tx laser phase noise contributions are not identical in all received sub-bands, but are relatively delayed across sub-bands, as indicated by the argument of the Tx-LPN expression ϕ0Tx-LPN[k−τ(i)]. Indeed, while Tx-LPN is common to all transmitted sub-bands (by the argument made above for the LO), nevertheless, in the course of optical propagation, due to CD, each of the sub-bands accumulates a different delay, τ(i) for the i-th sub-band. Thus, simultaneous arrivals at the Rx in different sub-bands at discrete-time k correspond to relatively delayed temporal samples of the Tx laser phase noise. It follows that there is dependence of the Tx-LPN contributions at the Rx on the index i via the delays τ(i). Also notice that the CFO and pilot frequency offsets and the Tx and Rx phase dithers are all common phase contributions. Finally, we may verify that by setting i=0 in Eq. (15) we retrieve Eq. (14) (14), as we have implicitly assumed without loss of generality that τ(0)=0, as the zeroth sub-band delay is taken as the time origin at the Rx.
Another key term which to a very good approximation constitutes common phase equally affecting all the sub-bands is the inter-sub-band XPM which depends on the total power of all ‘the other’ sub-bands (i.e. for the effect on the i-th sub-band, the powers of all the sub-bands with an index different than i should be summed up. The total power affecting sub-band i may be expressed as the total power of the full WDM signal minus the power of the i-th sub-band. As the WDM signal includes multiple channels each having multiple sub-bands, the power of a single sub-band may be of the order of 1% of the total power of the WDM signal, thus the total power affecting each of the sub-bands is nearly constant. This justifies why we were able to set the index of the XPM phase noise term to 0.
Finally notice that it is just the non-linear phase noise contributions which have a dependence on polarization.
We now describe the functionality and operation of the ‘pilot offset+optional dither phase cancellation stage’. The role of this module, operating in the angular domain, after the ‘angle extract’ operation, is to cancel out the phase term θpk out of ψ0[k] (Eq. (14)) yielding the reduced pilot phase output, labeled by a P subscript:
ψ0P[k]=ψ0[k]−θpk=θCFOk+ΣlβdTx[l] sin(θdTx[l]k+φdTx[l])+ϕ0X/Y-FWM[k]+ϕ0X/Y-SPM[k]+ϕ0X/Y-XPM[k]+ϕ0X/Y-ASE[k]+ϕ0Tx-LPN[k]+ϕ0Rx-LPN[k] (16)
The phase ramp term θpk associated with the carrier frequency offset, which was canceled out is deterministically known, as the phase offset θp is determined by the pilot sub-carrier index, counted relative to the DC sub-carrier, as follows:
θp=(sparse pilot sub-carrier index)·2π/(IFFT size)
As for the sinusoidal phase modulation dither terms, those also have known frequencies as the dither is deterministically induced (but unknown amplitudes and phases due to unknown propagation losses and delays) therefore a pair of narrowband offset PLLs may readily lock onto these phases and derive accurate estimates of these phase domain sinusoidal waveforms. Therefore the entire signal θpk+ΣlβdTx[l] sin(θdTx[l]k+φdTx[l]) is readily estimated and is subtracted out as indicated in Eq. (16), yielding the reduced common phase estimate ϕ0P[k] to be passed on to each of the sub-bands for demodulation, and also to be input into the CFO PLLs stage of the mid-sub-band PROC. The processing in the i-th sub-band has already been described in
The processed sub-band signals {tilde under (R)}k(±i) are both multiplied by the conjugate of the exponentiated feedforward common phase estimate ejφ
{tilde under (R)}k(i)e−jφ
The demodulation then amounts to subtracting the reduced phase of the 0th sub-band as generated by the mid-sub-band PROC (Eq. (16)) out of the total phase of the i-th sub-band (Eq. (15)). Performing the subtraction of the two equations term by term, it is apparent that several terms cancel out, yielding the following residual phase difference in the i-th output of the CFO(+PN) COMP stage:
Δψi[k]≡ψi[i]−ψ0P[k]=ΔϕiX/Y-FWM[k]+ΔϕiX/Y-SPM[k]+ΔϕiX/Y-ASE[k]+ΔϕiTx-LPN[k] (18)
where ΔϕiX/Y-FWM[k]≡φiX/Y-FWM[k]−ϕ0X/Y-FWM[k];
ΔϕiX/Y-SPM[k]≡φiX/Y-SPM[k]−ϕ0X/Y-SPM[k];
ΔϕiX/Y-ASE[k]≡φiX/Y-ASE[k]−ϕ0X/Y-ASE[k];
ΔϕiTx-LPN[k]≡φiTx-LPN[k−τ(i)]−ϕ0Tx-LPN[k]
The terms which have been cancelled out are the phases common to all sub-bands, namely the CFO, the phase dithers, the XPM phase noise and the Rx LO laser phase noise. Significantly, the cancellation of the Rx LO LPN implies that LO equalization enhanced phase noise (LO-EEPN) is gone. LO-EEPN is a type of phase noise impairment in conventional coherent receivers, whereby the CD group delay equalization in the Rx generates extra phase noise due to the LO LPN being delayed differently at each frequency in the process of CD equalization. In principle, LO-EEPN is also present in the sub-band processing based Rx-s, whenever there is LO PN, as the CTO varies per sub-band, hence the CTO comp induces different delays onto the Rx LO LPN (notice that the Tx LPN experiences no net enhancement due to CD equalization, since the various transmitted sub-bands are relatively delayed in the fiber link but they are then relatively delayed in the opposite sense in the CTO COMP modules of each sub-band, such that the net delay on the Tx phase noise is zero). Now, by virtue of using the sparse pilot tone, the Rx LO LPN is nearly ideally cancelled hence the EEPN impairment is essentially absent from our Rx-s.
The four phase noise terms which are not cancelled out, and may be even enhanced, are the FWM, the SPM, the ASE and the Tx-LPN. As the two FWM terms in sub-bands 0 and i are correlated, their substraction ΔϕiX/Y-FWM[k] may result in either smaller or larger variance of the fluctuations, depending on their coefficient of correlation. Additional FWM cancellation measures, outside the scope of this application may be applied. As for the SPM terms, these are uncorrelated, as each is generated solely from subcarriers in its own band, hence the fluctuation variance of ΔϕiX/Y-SPM[k] is doubled due to the pilot phase subtraction, however the SPM contribution may be small to begin with, hence this term might not significant.
As for the two ASE terms subtracted to yield ΔϕiX/Y-ASE[k], these are also uncorrelated, however by selecting the power of the sub-carrier to be at least as large as the average power of each data sub-band, this term may be made quite negligible. E.g., if the sparse pilot power is taken to be precisely equal to the average power of each data sub-band, then that implies that the sparse pilot amplitude be √{square root over (N)} larger than the RMS amplitude of the data sub-carriers, where N is the number of sub-carriers per sub-band, e.g. N=64. Therefore, for OFDM detection, the linear phase noise variance due to the additive circular Gaussian noise added to the sub-carrier phasors, is N times smaller for the sparse pilot than for the i-th data sub-band, and the ASE-induced LPN enhancement factor is just 1+N−1, which is negligible for a large value of N.
The remaining Tx-LPN term ΔϕiTx-LPN[k]=φiTx-LPN[k−τ(i)]−ϕ0Tx-LPN[k] has an interesting behavior due to the various delays. To see this we substitute Eq. (18) into Eq. (17), yielding the following input into the CTO COMP module,
|{tilde under (R)}k(i)|ejΔψ
therefore the CTO COMP output is a time-advance of this signal (substituting k+τ(i) for k) as follows
|{tilde under (R)}k+τ
The time-advance has no significant impact on the four terms on the RHS of the last equation but it does wield an interesting effect onto the last Tx-LPN term, which is singled out below:
ejΔϕ
It is now seen that the Tx-LPN of the i-th sub-band has been time-realigned but now the subtracted Tx-LPN of the mid-sub-band, ϕ0Tx-LPN[k+τ(i)] has been advanced (negative delayed) in time. The overall phase noise is now the sum of these two contributions which tend to be more uncorrelated the longer the delay τ(i) is, i.e., the higher the absolute value |i| is. Thus more remoted from the center frequency sub-bands tend to generated more phase noise.
In fact, this phase noise impairment is the dual of the LO-EEPN phase noise impairment described above, and may be referred to as Tx-EEPN phase noise impairment. So, we see that while we have suppressed LO PN altogether, we have replaced the LO-EEPN impairment in our system by an analogous Tx-EEPN impairment. Now the LO phase noise has been eliminated altogether, but in a sense the zeroth sub-band plays the role previously played by the LO, as the zeroth sub-band is used to demodulate the received signal, which contains Tx phase noise (but the zeroth sub-band contains phase noise itself).
The residual Tx-EEPN phase noise of Eq. (19) in the worst case (most extreme sub-band) should be qualitatively compared with the overall LO-EEPN phase noise in a “full-band” coherent Rx (i.e., without using FB based sub-banding).
The residual Tx-EEPN phase noise may be partially mitigated in our MSB (DFT-S) ODM by the carrier recovery system further downstream, just prior to the decision stage in each sub-band Rx. In our SC-MSBE Rx (
The method disclosed here of using a sparse pilot tone in the mid-sub-band in order to demodulate the other sub-bands with it and thus cancel impairments such as the CFO and the non-linear phase noise, will also work if the sparse pilot tone is transmitted in another sub-band with index i≠0 such that its center frequency does not coincide with the center frequency of the whole channel,
However, when using a sub-band other than the mid-sub-band for pilot transmission, the performance might be somewhat degraded as the maximal spectral distance between the pilot and each of the other sub-bands is increased, and the sub-band farthest away may have less correlation between its phase and the phase of the sub-carrier.
Finally, the method disclosed here of using a sparse pilot tone in the mid-sub-band in order to demodulate the other sub-bands at the output of the OS analysis FB should be compared with prior art usage of the sparse pilot tone to compensate laser phase noise and non-linear phase noise, as described in references [REF. A, REF. B] and other references there in. REF. A refers to Randel, S.; Adhikari, S.; Jansen, S. L.; “Analysis of RF-Pilot-Based Phase Noise Compensation for Coherent Optical OFDM Systems,” Photonics Technology Letters, IEEE, vol. 22, no. 17, pp. 1288-1290, Sep. 1, 2010. REF. B refers to Liang B. Y. Du and Arthur J. Lowery, “Pilot-based XPM nonlinearity compensator for CO-OFDM systems,”, Optics Express, Vol. 19, pp. B862-B867 (2011).
In prior art, wherein FB based sub-banding is not used, the sparse pilot tone (i.e., a pilot tone with null guardbands on either side) is extracted by means of a bandpass filter and used to demodulate the entire broadband optical channel. Thus a relatively narrowband pilot band is extracted out of the full bandwidth of the channel. The bandpass filter (BPF) and the demodulator should in principle operate at a sampling rate R equal to or higher than the bandwidth B of the entire channel. In order to prevent loss of spectral efficiency, the pilot extracting BPF should be spectrally sharp, and relatively narrowband, but not too narrow, so as to capture the bandwidth of the disturbance imparted upon the pilot tone in the course of optical propagation.
Initially, a pilot band of the order of tens of MHz was suggested in [Ref. A], aimed at mitigating laser phase noise, however in later works [Ref. B] the sparse pilot band was increased to the order of magnitude of GHz in order to capture the non-linear phase noise spectral width around the pilot.
The problem with the prior work approach is the prohibitive complexity incurred in digitally implementing a sharp narrowband filter of bandwidth of the order of B/M, where the bandwidth ratio M is typically in the range 10 . . . 20, operating at the full channel sampling rate R≥B. Because of the sharpness of the pilot extracting filter, a large number of taps, P, will be required in this filter, operating at full rate R, thus the number of complex taps per second is PR.
It turns out that the complexity of this filter is much larger than the complexity of the polyphase filters array of a reference CS FB which has its number of sub-bands set to be equal to M, and even larger than the complexity of an OS FB.
Indeed, comparing first with the CS FB complexity, a CS FB has M polyphase components each with P/M taps at the output of the S/P, i.e., operating at the rate R/M. Thus, the total number of complex multiplications per second in the polyphase array of the filter bank plus the IFFT is
This is the complexity of the CS FB to be compared with the complexity PR for the pilot-extracting BPF, i.e. the complexities ratio is
and this ratio is much less than unity for typical values of M and P which can be required to be of the order of tens of taps.
Notice that the complexity of an OS analysis FB will be even smaller than that of the CS analysis FB, thus in turn the OS analysis FB is substantially less complex than the pilot extracting FB. Moreover, in our solution for pilot extraction we just use one of the sub-bands of the OS analysis FB which is there anyway to perform the main function of channel processing for detection, therefore there is no incremental extra complexity penalty in extracting the sparse pilot in our approach (as the FB hardware used to do it is already there). It follows that usage of the OS filter bank entails a substantial complexity advantage over prior art extraction of the sparse pilot tone.
This completes the description of the mid-sub-band processor of
The invention may also be implemented in a computer program for running on a computer system, at least including code portions for performing steps of a method according to the invention when run on a programmable apparatus, such as a computer system or enabling a programmable apparatus to perform functions of a device or system according to the invention.
A computer program is a list of instructions such as a particular application program and/or an operating system. The computer program may for instance include one or more of: a subroutine, a function, a procedure, an object method, an object implementation, an executable application, an applet, a servlet, a source code, an object code, a shared library/dynamic load library and/or other sequence of instructions designed for execution on a computer system.
The computer program may be stored internally on a non-transitory computer readable medium. All or some of the computer program may be provided on computer readable media permanently, removably or remotely coupled to an information processing system. The computer readable media may include, for example and without limitation, any number of the following: magnetic storage media including disk and tape storage media; optical storage media such as compact disk media (e.g., CD-ROM, CD-R, etc.) and digital video disk storage media; nonvolatile memory storage media including semiconductor-based memory units such as FLASH memory, EEPROM, EPROM, ROM; ferromagnetic digital memories; MRAM; volatile storage media including registers, buffers or caches, main memory, RAM, etc.
A computer process typically includes an executing (running) program or portion of a program, current program values and state information, and the resources used by the operating system to manage the execution of the process. An operating system (OS) is the software that manages the sharing of the resources of a computer and provides programmers with an interface used to access those resources. An operating system processes system data and user input, and responds by allocating and managing tasks and internal system resources as a service to users and programs of the system.
The computer system may for instance include at least one processing unit, associated memory and a number of input/output (I/O) devices. When executing the computer program, the computer system processes information according to the computer program and produces resultant output information via I/O devices.
In the foregoing specification, the invention has been described with reference to specific examples of embodiments of the invention. It will, however, be evident that various modifications and changes may be made therein without departing from the broader spirit and scope of the invention as set forth in the appended claims.
Moreover, the terms “front,” “back,” “top,” “bottom,” “over,” “under” and the like in the description and in the claims, if any, are used for descriptive purposes and not necessarily for describing permanent relative positions. It is understood that the terms so used are interchangeable under appropriate circumstances such that the embodiments of the invention described herein are, for example, capable of operation in other orientations than those illustrated or otherwise described herein.
The connections as discussed herein may be any type of connection suitable to transfer signals from or to the respective nodes, units or devices, for example via intermediate devices. Accordingly, unless implied or stated otherwise, the connections may for example be direct connections or indirect connections. The connections may be illustrated or described in reference to being a single connection, a plurality of connections, unidirectional connections, or bidirectional connections. However, different embodiments may vary the implementation of the connections. For example, separate unidirectional connections may be used rather than bidirectional connections and vice versa. Also, plurality of connections may be replaced with a single connection that transfers multiple signals serially or in a time multiplexed manner. Likewise, single connections carrying multiple signals may be separated out into various different connections carrying subsets of these signals. Therefore, many options exist for transferring signals.
Although specific conductivity types or polarity of potentials have been described in the examples, it will be appreciated that conductivity types and polarities of potentials may be reversed.
Each signal described herein may be designed as positive or negative logic. In the case of a negative logic signal, the signal is active low where the logically true state corresponds to a logic level zero. In the case of a positive logic signal, the signal is active high where the logically true state corresponds to a logic level one. Note that any of the signals described herein can be designed as either negative or positive logic signals. Therefore, in alternate embodiments, those signals described as positive logic signals may be implemented as negative logic signals, and those signals described as negative logic signals may be implemented as positive logic signals.
Furthermore, the terms “assert” or “set” and “negate” (or “deassert” or “clear”) are used herein when referring to the rendering of a signal, status bit, or similar apparatus into its logically true or logically false state, respectively. If the logically true state is a logic level one, the logically false state is a logic level zero. And if the logically true state is a logic level zero, the logically false state is a logic level one.
Those skilled in the art will recognize that the boundaries between logic blocks are merely illustrative and that alternative embodiments may merge logic blocks or circuit elements or impose an alternate decomposition of functionality upon various logic blocks or circuit elements. Thus, it is to be understood that the architectures depicted herein are merely exemplary, and that in fact many other architectures can be implemented which achieve the same functionality.
Any arrangement of components to achieve the same functionality is effectively “associated” such that the desired functionality is achieved. Hence, any two components herein combined to achieve a particular functionality can be seen as “associated with” each other such that the desired functionality is achieved, irrespective of architectures or intermedial components. Likewise, any two components so associated can also be viewed as being “operably connected,” or “operably coupled,” to each other to achieve the desired functionality.
Furthermore, those skilled in the art will recognize that boundaries between the above described operations merely illustrative. The multiple operations may be combined into a single operation, a single operation may be distributed in additional operations and operations may be executed at least partially overlapping in time. Moreover, alternative embodiments may include multiple instances of a particular operation, and the order of operations may be altered in various other embodiments.
Also for example, in one embodiment, the illustrated examples may be implemented as circuitry located on a single integrated circuit or within a same device. Alternatively, the examples may be implemented as any number of separate integrated circuits or separate devices interconnected with each other in a suitable manner
Also for example, the examples, or portions thereof, may implemented as soft or code representations of physical circuitry or of logical representations convertible into physical circuitry, such as in a hardware description language of any appropriate type.
Also, the invention is not limited to physical devices or units implemented in non-programmable hardware but can also be applied in programmable devices or units able to perform the desired device functions by operating in accordance with suitable program code, such as mainframes, minicomputers, servers, workstations, personal computers, notepads, personal digital assistants, electronic games, automotive and other embedded systems, cell phones and various other wireless devices, commonly denoted in this application as ‘computer systems’.
However, other modifications, variations and alternatives are also possible. The specifications and drawings are, accordingly, to be regarded in an illustrative rather than in a restrictive sense.
In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word ‘comprising’ does not exclude the presence of other elements or steps then those listed in a claim. Furthermore, the terms “a” or “an,” as used herein, are defined as one or more than one. Also, the use of introductory phrases such as “at least one” and “one or more” in the claims should not be construed to imply that the introduction of another claim element by the indefinite articles “a” or “an” limits any particular claim containing such introduced claim element to inventions containing only one such element, even when the same claim includes the introductory phrases “one or more” or “at least one” and indefinite articles such as “a” or “an.” The same holds true for the use of definite articles. Unless stated otherwise, terms such as “first” and “second” are used to arbitrarily distinguish between the elements such terms describe. Thus, these terms are not necessarily intended to indicate temporal or other prioritization of such elements. The mere fact that certain measures are recited in mutually different claims does not indicate that a combination of these measures cannot be used to advantage.
While certain features of the invention have been illustrated and described herein, many modifications, substitutions, changes, and equivalents will now occur to those of ordinary skill in the art. It is, therefore, to be understood that the appended claims are intended to cover all such modifications and changes as fall within the true spirit of the invention.
This application is a continuation of U.S. patent application Ser. No. 14/123,229 filing date Mar. 3, 2014 titled RECEIVER, TRANSMITTER AND A METHOD FOR DIGITAL MULTIPLE SUB-BAND PROCESSING which claims priority from provisional patent filing date Jun. 10, 2011, Ser. No. 61/495,533—both applications being incorporated herein in their entirety.
| Number | Name | Date | Kind |
|---|---|---|---|
| 20080316912 | Al Adnani | Dec 2008 | A1 |
| 20120155571 | Yang | Jun 2012 | A1 |
| Number | Date | Country | |
|---|---|---|---|
| 20170214558 A1 | Jul 2017 | US |
| Number | Date | Country | |
|---|---|---|---|
| 61495533 | Jun 2011 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 14123229 | US | |
| Child | 15358164 | US |
