Open Access
Add to BibSonomyAbstract
A simple and low cost method for wavelength control of economical random non-preselected independent ONU sources is shown to increase the number of users in an OFDMA-PON. The method is based on OLT monitoring and thermal tuning control; it has been validated through Monte-Carlo simulations and a probabilistic model. The minimum optical spectral gap between the ONUs wavelengths that guarantees a tolerable amount of optical beat interference has been determined through an experiment.
©2011 Optical Society of America
1. Introduction
Passive optical networks (PON) have become an interesting solution for deploying fibre to the home networks. Recently, orthogonal frequency division multiplexing (OFDM) modulation has been proposed for PONs combining time and frequency division in order to provide fine bandwidth granularity [1,2].
Generally, the upstream multipoint to point architecture of PONs is more challenging than the downstream. This is particularly true for OFDMA PONs where the subcarriers must be kept orthogonal to recover the data properly. Additionally, the optical network unit (ONU) light source has to be cost-effective for making it economically attractive. Depending on the optical source choice, two strategies are identified: colourless or coloured ONUs. In the first case, the ONU can reuse a signal sent from the optical line terminal (OLT) to modulate its data with reflective devices such as reflective semiconductor optical amplifiers (RSOA). This solution, however, faces problems related to the limited power gain of these reflective devices and Rayleigh backscattering effects in the fibre. Alternatively, a tuneable laser can also be employed as source in a colourless ONU, but its cost is currently excessive for access networks. Colored ONUs, instead, transmit with independent pre-selected light sources (like DFB) at specific wavelengths. The respective ONUs’ wavelength is chosen to avoid interference in upstream. The drawback relates to the large inventory and complex provisioning that operators would face, increasing the total cost of implementation.
In contrast, with non-preselected light sources, all ONUs can use the same type of laser module whose wavelength is random in a specified optical band. The process of adding a new wavelength to the PON involves trial and error lasers until a valid wavelength is obtained. This limits the number of users in the PON and the rejection probability of new lasers is high. Moreover, temperature fluctuations and phase noise effects also add to the random nature of these non-preselected sources. A wavelength control algorithm based on feedback between the ONU and OLT can reduce the randomness of these light sources [3]. This paper explores the feasibility to reduce the rejection probability of new colored non-preselected light source ONU with an algorithm for controlling the operative wavelength of the laser and extends the work presented in [4]. The ONU lasers are considered to follow a uniform probability and are thermally tuned by OLT control signals to displace slightly its wavelength and avoid overlaps with other ONUs. As a result, the users in the PON are increased and the laser rejection probability lowers rendering the PON implementation economically attractive.
Direct-detection is currently the most cost-effective and simple receiver technique [5]. When different ONUs transmit with their own light source and are detected with a single photodiode in the OLT, the upstream signal is affected with optical beat interference (OBI) [3,6,7]. OBI refers generally to the unwanted mixing products generated during the direct-detection process that can interfere in the OFDM data frequencies [3]. The minimum optical spectral gap between the ONUs’ wavelength is experimentally determined firstly and used in a Monte-Carlo simulation to calculate the number of lasers that can be placed in a given optical frequency band depending on the available tuning range. A probabilistic mathematical model is included and agrees with the simulation results which show that in an OFDM PON with non-preselected random optical sources in the ONUs, the laser rejection probability is reduced from 28% when no tuning is performed to less than 1% with the control algorithm.
2. Optical Beat Interference experimental results
In order to assess the minimum optical spectral gap between the ONUs needed to avoid interference, the experimental setup in Fig. 1 (left) was considered. Two users, ONU1 and ONU2, with different wavelengths transmit an OFDM signal consisting of 256 subcarriers with 4-PSK coded data. ONU1 used the first 128 subcarriers, while ONU2 used the last ones. Data was generated randomly for a total length of 218 bits. The Hermitian symmetry property was employed in order to get a real valued OFDM signal. The data was loaded to an arbitrary waveform generator (AWG) to get two analogue waveforms at 6.5 GSa/s, giving an effective bandwidth (BW) for ONU1 and ONU2 of 3.125 GHz each. Each electrical signal modulates the light source by means of a Mach-Zehnder modulator (MZM) biased at quadrature. The ONU1 light source was a DFB whose wavelength was left static at 1550.83 nm, while the ONU2 wavelength was swept from 0 to 0.8 nm with respect to the ONU1 emission wavelength. The total unmodulated laser linewidth was about 30 MHz. The optical OFDM signal then travelled through 6 km of fibre; it was detected with an avalanche photodiode (APD) and captured with a real-time 50 GSa/s oscilloscope. The signal was then post-processed in Matlab® performing the FFT and PSK decoding for measuring the bit error ratio (BER) of ONU1 data.
The results of the experiment in terms of BER against the spectral gap between the 2 ONUs sources are plotted in Fig. 2 , together with corresponding measured constellations and optical spectra. The received power was kept constant for all the points measured, taking as baseline the value when the gap was 0.8 nm. It can be observed that the BER is almost constant and below the FEC margin of 10−3 when the optical gap is more than 0.075 nm (9.375 GHz in C-band). It is a consistent result since it nearly corresponds to the sum of the modulated BWs of each ONU. This equals the situation where the two spectra have crossed each other completely and almost no overlap is present. Following the results of the experiment, in case more ONUs are transmitting simultaneously at the same bit rate, a spectral gap of at least 0.1nm would be needed between their emission wavelengths to reduce OBI and be properly detected.
Fig. 2 OBI BER against wavelength separation experimental results along with the constellations and optical spectra of the points indicated.
3. Laser wavelength distribution algorithm
We consider as the ONU upstream light source a conventional single-mode laser at any non-preselected wavelength; thus, will be described as following an uniform probability [8]. This random behavior could cause an overlap of the upstream signals; to reduce it, the emission frequency can be detuned by temperature, if OBI is detected.
3.1 Description
In general, the algorithm for an ONU addition would be the following: a) the customer acquires a standard ONU module based on low-cost DFB or VCSEL; its wavelength can be anywhere in the specified optical band (e.g. in C-band); b) the ONU is installed and added to the PON; c) the optical spectrum or performance of the upstream ONU signal is monitored and identified by the OLT; d) if interference or possible interference is detected, then a control signal is sent from the OLT to the ONU; e) the ONU source wavelength is tuned by varying the temperature, usually by heating, until a void slot is detected. If the whole tuning range is exhausted and no valid wavelength has been found, the laser is rejected and replaced. Figure 3 shows a flow diagram of this procedure.
Fig. 3 (left) Algorithm flow diagram, (right) example of user emission frequency allocation after the proposed λ-control algorithm. The trapezoids indicate each ONU BW, when the base of the trapezoid is larger it means it went through more tuning.
Temperature control allows for approximately 0.1 nm/ °C tuning for a DFB or a VCSEL [9]. It can be achieved by using a Peltier device or just by heating. Practically, the maximum forced temperature variation can be in the range of 1-2 nm. This depends on the installation conditions and on the specific design of the laser module. Figure 3 right depicts an example of applying the algorithm. Each trapezoid represents an ONU in its corresponding frequency and BW. The tuning, if present, is represented with the change in the left diagonal side, and it was restricted to higher frequencies for this figure. For instance, the ONU in light green initially has an emission frequency (~710GHz) that overlays the existing blue user, so it is tuned to the right to a free BW slot (~725GHz). Hence, in Fig. 3 right the longer the base of the trapezoid, the more tuning was applied to that ONU.
3.2 Theoretical approach
For ease of programming, the continuous grid is discretized into k bands with a width equal to the needed OBI spectral safe margin. The ONU wavelengths are considered to fall in any of these sub-bands with equal probability. A total of y wavelengths are generated following an uniform probability distribution, and the number of lasers that are found in each sub-band h is denoted as wh. OBI occurs when any of these wh is greater than one. We first analyze the case in which the number of sub-bands equals the total number of lasers to be allocated (y = k). For example, consider the total band is divided into 3 sub-bands (k = 3, h = {1,2,3}) and 3 lasers (y = 3) are generated. The only OBI-free case is having all ONUs in different sub-bands, therefore the probability of this situation can be modeled as following a multinomial distribution with all wh = 1:
In order to compute the probability that only x of the y ONUs transmit in different frequency, Eq. (1) has to be modified. Following the example, this refers to the case when we consider that only 2 ONUs (x=2) out of the 3 available (y=3) have enough separation between their emission frequency (w1=1, w2=2, and w3=0 and their permutations). Having these remarks on mind, Eq. (1) is modified as:
For a given y ONUs, we should consider all the possible situations that result in x active ONUs, indicated in the values of wh. As an example, if only 2 ONUs (x=2) transmit out of a total of 4 of them (y=4), the options are w1=w2=2, w3=w4=0 and w1=1, w2=3, w3=w4=0 with their corresponding permutations. As can be inferred, the sum of all wh should result in y: . The set that includes all these possibilities will be accounted for with the symbol W. Equation (2) when x≠y is adjusted as follows:
Finally, as the number of total available bands can be different that the y generated lasers (y≤k) the combinations of k in y are incorporated too. For example, having a grid where 4 users (k = 4) can transmit, but only 3 (y=3) are active. The OBI free probability is obtained as:
When the λ-control is applied, the probabilities are modified by the possibility of being tuned and displaced to an unoccupied band. The factors wk are changed to wk,tun to consider the available tuning. Following with the example of the previous paragraph if lasers can be tuned 1 band, then the possibility that the three of them interfere in the same sub-band h is avoided, and if the users can be tuned up to 2 bands, then no interference should happen (w1 = w2 = w3 = 1). Having these considerations in mind, only the denominator of Eq. (4) when x≠y changes and turns into:
In Eq. (5), tun refers to the number of slots that the laser can be displaced and needs to be solved recursively. We simplify it by using the concept of compositions from number theory in the summation: x numbers whose sum is equal to y [10]. As an approximation, we consider only the total number of compositions of a number along with the possible combinations inside each composition. The simplified result is plotted in Fig. 4 (left) for a total of 64 lasers. As noted, the amount of active lasers approaches the total lasers when the tuning increases, giving an acceptance ratio of around 99% with full tuning. These results were generated using Eq. (5) and provide a first estimation of the rejection ratios that can be expected, which are further validated through a Monte Carlo analysis next section.
Fig. 4 (left) Theoretical results from the simplified probability equation, (right) Monte Carlo results when 64 lasers are generated in a band with a maximum allocation for 64 users.
3.3 Simulation results
A Monte Carlo simulation following the algorithm of Fig. 3 was performed to determine the number of users that can be served in a practical PON system. As formerly observed from the experiment, a free optical spectral slot of the double of the total signal BW for each ONU can be considered a safe margin in order to avoid penalties caused by the user’s interference. The light emission frequency of the lasers is considered to follow a uniform probability distribution within a 1.6 THz band. Firstly, a total of 64 laser wavelengths was generated randomly and considered one at a time to fit in the spectral segment meeting the constraint of a minimum optical gap between consecutive lasers of 0.2 nm (25 GHz). With this spectral separation, the maximum number of lasers that can be accommodated within the total band is 64. If the laser frequency was overlapping with an existing one, then it was tuned upwards. We considered tuning ranges from 0 (meaning no tuning) up to 250 GHz (about 2 nm range thermal tuning with 20 °C). In case that tuning puts the laser out of the range or it exceeds the tuning limit available, then the laser is rejected and should be replaced by another. Each point was simulated 100 times and the mean values were considered.
The results of the simulation in terms of number of active and rejected lasers with respect to the laser thermal tuning range are illustrated in Fig. 4 right. As it can be noted, as the tuning range is increased, the overlapping probability strongly lowers. When no tuning is in place, 72% of the 64 lasers can be accommodated, but if the algorithm is applied with 2nm thermal tuning, then 99% of them can be active in the PON. This reduces the last entry laser rejection rate from 28% to about 1%. The theoretical results plotted in Fig. 4 left provided similar rejection rate (~1%) when the tuning was applied, indicating that both approaches converge. However, when ONUs are not tuned, the probabilistic model expected fewer rejected laser (~13%). This difference is caused mainly by the simplification approach used to compute the probability of Eq. (5).
As a further evaluation of the algorithm, the uplink bitrate was reduced by half, and the OBI safe margin was then set to 0.1 nm, allowing to allocate a total of 128 lasers. The results are plotted in Fig. 5 (left). In this case, the rejection probability of the last laser in the PON decreases from 63% to less than 1%. If the upstream bitrate is further lowered by another half, a total of 256 lasers could be placed. For this situation, a total of 32, 64, 128 and 256 lasers were generated and their probability of being active or rejected is depicted in Fig. 5 right. In all cases, the replacement ratio of the final laser was reduced to less than 1%, and the fewer lasers were targeted, the less tuning was required. These results show that the λ-control algorithm proposed can turn the replacement cost of ONUs with random non-preselected light sources affordable for operators.
Fig. 5 (left) Monte Carlo results when 128 lasers are generated in a band with a maximum allocation of 128 users for several tuning BW; (right) Monte Carlo results when 32, 64, 128, and 256 lasers are produced in a band with a maximum allocation of 256 users for several tuning BW.
More intelligence can be added to the algorithm, like adaptively setting an optimal allocation for all lasers, hence allowing for more random-λ ONUs to operate in the PON, consequently improving the overall BW efficiency. Besides, this technique is not limited to OFDM and can also be employed with other modulation formats.
4. Conclusion
An algorithm for laser wavelength emission control with thermal tuning at ONU registration within the PON was analyzed in OFDMA type networks with low-cost non-preselected independent laser sources in ONUs. Even if the algorithm is simple, results show that in a 64-user PON, the mean probability that a laser will not interfere with another one increases from 72% up to 99% with a thermal tuning range of 2nm. If required, more intelligence can be added to the algorithm in order to find the optimum allocation of ONUs wavelengths and also monitor and control the possible wavelength fluctuations in an operative environment. This provides an approach for ONUs upstream transmission in support of cost effective OFDMA networks; apart from other modulation formats, in which this technique can be applied too.
Acknowledgments
This work was supported by the European project ACCORDANCE and CONACYT grant 149826.
References and links
1. W. Wei, L. Zong, and D. Qian, “Wavelength-based sub-carrier multiplexing and grooming for optical networks bandwidth virtualization,” in Proceedings OFC 2008, paper PDP35 (2008).
2. D. Qian, N. Cvijetic, J. Hu, and T. Wang, “A novel OFDMA-PON architecture with source-free ONUs for next-generation optical access networks,” IEEE Photon. Technol. Lett. 21(17), 1265–1267 (2009). [CrossRef]
3. S. Soerensen, “Optical beat noise suppression and power equalization in subcarrier multiple access passive optical networks by downstream feedback,” J. Lightwave Technol. 18(10), 1337–1347 (2000). [CrossRef]
4. I. Cano, M. C. Santos, V. Polo, and J. Prat, “Dimensioning of OFDMA PON with non-preselected independent ONUs sources and wavelength-control,” in Proceedings ECOC 2011, paper Tu.5.C.2 (2011).
5. S. L. Jansen, B. Spinnler, I. Morita, S. Randel, and H. Tanaka, “100 GbE: QPSK versus OFDM,” Opt. Fiber Technol. 15(5-6), 407–413 (2009). [CrossRef]
6. C. Desem, “Optical interference in subcarrier multiplexed systems with multiple optical carriers,” IEEE J. Sel. Areas Comm. 8(7), 1290–1295 (1990). [CrossRef]
7. C. H. Chang, “Interference of multiple optical carriers in subcarrier-multiplexed systems,” IEEE Photon. Technol. Lett. 5(7), 848–850 (1993). [CrossRef]
8. A. Papoulis, Probability, Random Variables and Stochastic Processes (McGraw-Hill, New York, 1965).
9. S. Uchiyama, N. Yokouchi, and T. Ninomiya, “Continuous-wave operation up to 36°C of 1.3-μm GaInAsP-InP vertical-cavity surface-emitting lasers,” IEEE Photon. Technol. Lett. 9(2), 141–142 (1997). [CrossRef]
10. S. Heubach and T. Mansour, Combinatorics of Compositions and Words (CRC Press, Boca Raton, FL, 2009).
