This invention relates generally to optical communication, and more particularly to coherent fiber optic communication.
A desire to increase both the data rate and transmission distance of optical communication links has prompted the use of coherent signaling. Conventionally, optical communication systems have relied on the use of simple signaling methods to encode data bits onto an optical carrier. The most common signaling method is intensity modulation, in which a laser is gated to allow high intensity light to enter a fiber optical cable when a ‘1’ bit is to be transmitted and low intensity light when a ‘0’ bit is transmitted. This is a form of on-off keying and has the advantage that it is easily demodulated by a photo-diode and an appropriate threshold.
The main drawback of on-off keying is that the bandwidth efficiency is low, due to the fact that information is only transmitted in a single dimension. Coherent signaling techniques, on the other hand, allow for the transmission of multidimensional signals, by modulating both the intensity and the phase of the laser light.
The dimensional signal is passed through an optical channel, 104, and is detected and demodulated in a receiver 105. A coherent receiver includes a second laser light source (receive Local Oscilator RXLO) 101′, an optical hybrid demodulator and photo detectors (termed a “coherent detector”), 106, and an electrical digital receiver 107.
Several impairments affect the performance of the optical communication link, including the effect of non-ideal lasers. An ideal laser output can be expressed as
Ecw(t)=√{square root over (Ps)} exp(jωst+θs)es,
where Ecw(t) is an optical carrier, t is time √{square root over (Ps)} is the an amplitude, ωs is a frequency in radians per second, θs is an initial phase, and es is a polarization of the optical carrier. Deviations from the above ideal laser output are caused by spontaneous emitted photons, which cause intensity and phase fluctuations and result in a laser output that can be expressed as
Ecw(t)=√{square root over (Ps+δP(t))}exp(jωst+θs+θns(t))es,
where δP(t) and θns(t) represent intensity and phase noise processes, respectively.
Because these processes are due to the spontaneous emissions, they are reasonably modeled as Gaussian distributed random processes. The effect of the phase noise is to broaden a power spectral density of the optical field. Thus, the laser output is no longer confined to a single frequency tone, but typically has a Lorentzian-shaped spectrum. The full-width half maximum (FWHM) bandwidth is normally termed the “laser linewidth.” The phase noise, θns(t), or equivalently the laser linewidth, is particularly troublesome for coherent systems in that the phase of the carrier Ecw(t) is needed to coherently demodulate any two dimensional modulation format.
The random carrier phase θns(t) impairs carrier recovery processes, and generally the larger the laser linewidth, the more difficult it is for the receiver to track the carrier phase changes. The coherent receiver makes use of an additional laser 101′ to generate a local version of the optical carrier for mixing with the received signal. The non-ideal linewidth of this laser also compound the effects of the transmit laser's linewidth. Typical linewidth requirements for lasers in coherent systems are on the order of tens to hundreds of kHz, whereas direct detection or differentially Quadrature Phase Shift Keying (QPSK) can greatly reduce the linewidth requirements to tens of MHz because the signal formats are less sensitive to carrier phase changes. Studies show that a QPSK system requires maximum linewidth of 250 KHz, and a 16QAM systems requires 6.9 KHz linewidth.
In addition to laser linewidth, the received signal also experiences linear dispersive effects such as chromatic dispersion, as well as nonlinear effects, such as fourwave mixing as the signal traverses the optical fiber. Typical receivers make use of electronic signal processing to reduce the dispersive and non-linear effects. Thus, it is desirable to compensate for laser linewidth effects using electronic signal processing techniques as well.
In addition to the zero-mean phase noise, carrier frequency offset (CFO) ΔF may exist between two oscillators due to the temperature variation, aging and other slow effects. CFO is generally very slow with regard to the symbol rate, and therefore can be considered constant over a number of transmitted symbols. CFO adds additional phase error.
The non-zero linewidth Δv of the local oscillator (LO) results in a phase noise from one received symbol r(n) to the next symbol r(n+1) is represented by
δθk=θk−θk−1=Ts[(0,(2πΔv)2)+ΔF],
where Ts is the symbol duration and (0, σ2) denotes a zero mean normal distribution with variance, σ2. The method described in this invention, however, does not require the phase noise distribution to be normal, or zero-mean.
The phase error of the kth symbol can be expressed as
To limit the phase error, most conventional systems use LOs with very narrow linewidth. Because the requirement of the linewidth is also related to the modulation scheme, modulations with higher spectrum efficiency requires a much tighter LO linewidth.
To achieve the required performance, conventional coherent optical systems use lasers with very narrow linewidth as local oscillators, typically external cavity distributed feedback (DFB) lasers. These lasers have the linewidth Δv in the tens of kHz range, but are very expensive, have larger form factor, and higher power consumption.
Many other laser sources such as DFB lasers and vertical-cavity surface-emitting laser (VCSELs) are more cost and power efficient. However, the strict linewidth requirement prohibits the use of these inexpensive, and more energy efficient lasers in coherent systems, as the receiver performance degrades rapidly with increasing linewidth. Most of the DFB lasers, VCSELs and tunable lasers have linewidth that is in the range of MHz. Conventional current coherent systems will not function with adequate performance under such large linewidth.
Therefore, it is desirable to design coherent systems with relaxed linewidth tolerance to allow lower cost, or tunable light sources to be used in the coherent system. This will significantly lower the overall system cost and energy consumption. Tunable lasers also provide flexibility and reconfigurability to coherent systems.
The embodiments of the invention provide a coherent receiver and method that can compensate for phase error induced by non-zero linewidth and carrier frequency offset of local oscillators (LO), i.e., lasers, and effectively increases the linewidth tolerance. The method can improve the performance of the transmission at a given linewidth, or achieve the same performance with a larger linewidth.
The receiver estimates the phase error of the received blocks of symbols with the assistance of a channel decoder, and compensates the phase error iteratively. Each block of received symbols is equalized and phase compensated using an initial estimate of the phase.
The equalized and phase compensated symbol block is then fed into a decoder to produce information bits. The decoded information bits are used to regenerate the transmitted symbols.
A phase estimator compares the received symbols and the regenerated transmitted symbols, and outputs an estimated average phase error. The average phase error estimate is used to apply phase compensation on the next block, which is read in by the decoder to decode additional information bits. This compensation process iteratively continues until the entire received signal is compensated and decoded.
This compensation also produces an updated phase compensated symbol stream. The receiver can carry out this method once, or in multiple iterations. Each iteration refines the phase estimation for the block and updates the phase compensate stream to be used by the next iteration. The iterations terminate when a termination condition is satisfied, e.g., a fixed number of iterations has been reached, or the decoded information bit sequence converges.
In this description, the following variables are defined and used:
The symbol index is in the parenthesis, e.g., r(n). A superscript is used as step index and a subscript is used for iteration index. These are sometimes omitted for clarity when there is no cause for confusion. A bold font indicates the variable is a vector/matrix. A circumflex (hat) (â) is used to indicate that the variable a is an estimate.
At the start, indices l and k are initialized. Then in each step, the receiver:
An iteration is completed at the end 630 of the symbol block. The entire process terminates 640 when a termination condition is satisfied, e.g., a predetermined maximum number of iteration is reached, or the decoded information bit sequence converges. The receiver then outputs the decoded information bit sequence of the final iteration.
The equalizer 1001 compensates for the effects from the non-ideal fiber optic channels, such as chromatic dispersion, polarization mode dispersion, and non-linear effects, etc. The equalized blocks of received symbol sequences are denoted as r0=[r0(1), r0(1), . . . , r0(N)] 301, where r0(n) is the nth symbol and N is the total number of symbols in the sequence.
The phase estimator 1004 compares two blocks of symbols and estimates the average phase error θ between the blocks. The estimator can also generate a probability function representing the phase error, such as the log likelihood of the estimate L({circumflex over (θ)}).
Delay block (DLY) 1005 is used to match input symbols r(n) and s(n) in the time domain for the phase estimator.
The FCMP 1002 and the backward phase compensator 1011 compensate the phase rotation of a block of input symbols rk with an estimated phase error θk to produce a block of new symbols.
The de-interleaver 1007 is used before the decoder when the transmitter has an interleaver after the encoded data. An interleaver 1008, implemented in pair with the de-interleaver 1007, re-orders the regenerated symbol to match the order of the received symbols.
The error correction decoder 1003 receives the phase compensated symbol sequence blocks u, and outputs a decoded information bit sequence {circumflex over (d)}. If a soft output decoder is used, the decoder also produces the soft likelihood of the information bit sequence L({circumflex over (d)}) 306.
The FEC encoder 1006 and the interleaver 1008 regenerate an estimated transmitted symbol sequences ŝl from an estimated information bit sequence {circumflex over (d)} in the lth iteration.
Details of Method and Receiver
More details of the method and the receiver are described as follows:
The initial phase {circumflex over (θ)}0 (0) of the first received symbol is known and the received symbol sequence r0 has already been phase rotated, such that θ0(0)≈0. This is reasonable, and such estimation can be done, e.g., by evaluating the phase of known training preamble symbols preceding the data symbol stream. The original equalized received symbol sequence is denoted as
r0=[r0(1),r0(2) . . . ,r0(N)],
where r0(n) is the nth symbol and N is the total number of symbols.
In each step 601, the receiver estimates the average phase error of a block of P symbols. Each iteration 602 has
steps.
In the kth step, the average phase error of the kth block θk is estimated using the previous symbols in the (k+q−1)th blocks. The number of the additional symbol blocks needed by the decoder to decode information bits for block k is q−1, e.g. the decoding delay.
At the beginning of each iteration the receiver reads in q blocks of symbol, r1, r2, . . . , rq, de-interleaves the symbols. Then, the decoder estimates the information bits {circumflex over (d)}1 304. Note that the first q blocks of the forward phase compensated sequence u 303 are the same as r, i.e., ui=ri for i≦q;
The receiver encodes {circumflex over (d)}1 and interleaves the re-encoded symbols to produce an estimate of the P transmitted symbols in the first block, ŝ1. Then, the receiver estimates the average phase error of the first block 302 using the first block of received symbols r1 and the 1st block of estimated transmitted symbols ŝ1.
After, the first {circumflex over (θ)}1 become available, the receiver uses the estimate to compensate the phase of the (q+1)th block of the symbols rq+1. For example, the phase compensation can be performed as uq+1(n)=rq+1(n) exp(−j{circumflex over (θ)}1), for all symbols in the (q+1)th block. The input to the de-interleaver/decoder is u.
Next, the newly generated block uq+1 is fed to the de-interleaver and decoder to generate an information bit block {circumflex over (d)}2 and regenerate ŝ2 and eventually the average phase error estimate of the next block {circumflex over (θ)}2.
The next symbol block is phase compensated to be used for the following step. After the phase estimation for the kth block is complete, the receiver proceeds to the (k+1)th block. This step repeats until all blocks have been processed in the first iteration.
For most of the blocks, (q−1) additional blocks are required to decode one block of data properly. The last blocks of the sequence generally do not require additional blocks for decoding when codes are properly terminated e.g., convolutional codes typically are terminated to a known (zero) state by inserting tail bits. This allows the process to continue until it reaches the end of the sequence and produces phase estimate for all K blocks, instead of just K−q+1 blocks.
The backward phase compensator generates a new symbol sequence rl for the next (l+1)th iteration. The new symbols sequence is derived by directly compensating symbol block rk 315 with an estimated average phase error {circumflex over (θ)}k
rlk(n)=rl-1k(n)exp(−j2π{circumflex over (θ)}lk), for all kε[0,K]; or
rlk=rl-1ke−j2π{circumflex over (θ)}
In the kth step, the decoder 1003 outputs kth information bit sequence {circumflex over (d)}k given partial symbol sequence u1, u2, . . . uk+q. For example, if a maximum likelihood (ML) decoder is used, the decoded data are
{circumflex over (d)}k=argmax Pr(dk|{circumflex over (d)}1:k−1;u1:k+q),
where the function arg max returns the argument of the maximum.
Other types of decoders such as a maximum a posteriori (MAP) decoder or a linear decoder can also be used. Although soft input soft output (SISO) decoders offer better phase estimation performance, the receiver described herein invention does not require SISO decoders. Extra symbols are sometimes needed for correct decoding and as a result, additional symbol blocks Uk+1, . . . , uk+q−1 are read in by the de-interleaver and decoder.
The phase estimator 1004 determines the average phase rotation of the input symbol sequence block rl-1k 315 by comparing it with the estimated transmitted symbol block ŝlk 305. Assuming Gaussian distributed noise, the ML estimator is
or a linear estimator
where ∠(,) is the operator that computes the angle of two complex numbers, wn is a weight coefficient that can be determined based on noise variance σ(n), amplitude of r(n) and/or L(s(n)). In the case of equal weights, the estimator can be further simplified to
The forward phase compensator 1002 and backward phase compensator 1011 can both be implemented by rotating the symbols within block k+q by an angle of −{circumflex over (θ)}lk corresponding to the average phase error, given as
uk+q(n)=rl-1k+q(n)exp(−j2π{circumflex over (θ)}lk), for all kε[1,K−q+1], and
rlk(n)=rl-1(n)exp(−j2π{circumflex over (θ)}lk), for all kε[1,K].
The forward phase compensator 1002 and backward phase compensator 1011 may also be implemented using ML equalizers, which produces the symbols uk+q. The log likelihood ratios (LLRs) of the symbols are as
L(u(n))=log(Pr└u(n)|rl-1(n),{circumflex over (θ)}lk┘), for all nεblock k+q;
and
L(rlk(n))=log(Pr└rlk(n)|rl-1k(n),{circumflex over (θ)}lk┘), for all nεblock k; and kε[1,K].
If both L(rl(n)) and L({circumflex over (θ)}k) are available, L(u(n)) can be approximated as
The receiver exits the outer-loop when the termination condition is satisfied. Some of the possible termination conditions include:
for all blocks.
Let t0 500 be the beginning of the received symbol sequence, ti the end of block i and the beginning of block i+1, tK is the end of the Kth block of the received sequence. In the example shown in
The first iteration starts at time t0. At time t2, symbol blocks r01 and r02 are received, d1 can therefore be decoded and {circumflex over (θ)}11 is determined. The forward compensator generates u3. The backward compensator generates r11. The final block r0K is available at tK 506 and both {circumflex over (θ)}1K−1 and {circumflex over (θ)}1K can be determined after tK 506, and so forth. The 2nd iteration therefore can start at t2 502. Block r11=r01e−j2π{circumflex over (θ)}
Note that r02 and {circumflex over (θ)}12 are actually available at as early as t3 503 and therefore the 2nd iteration can start as early as at t3 503.
The receiver described herein can effectively remove a significant portion of the phase error and therefore enable the system to have much higher tolerance to the linewidth of local oscillators at both the transmitter and the receiver. The receiver can improve the overall performance of the transmission or/and allows system be built using local oscillators with wider linewidth to reduce the cost.
Although the invention has been described with reference to certain preferred embodiments, it is to be understood that various other adaptations and modifications can be made within the spirit and scope of the invention. Therefore, it is the object of the append claims to cover all such variations and modifications as come within the true spirit and scope of the invention.
Number | Name | Date | Kind |
---|---|---|---|
6625231 | Shen | Sep 2003 | B1 |
7724836 | Hartmann et al. | May 2010 | B2 |
7792220 | Alderson et al. | Sep 2010 | B2 |
8045634 | Zhang | Oct 2011 | B2 |
8116367 | Carrer et al. | Feb 2012 | B2 |
20110129234 | Duan et al. | Jun 2011 | A1 |
Number | Date | Country | |
---|---|---|---|
20110129234 A1 | Jun 2011 | US |