OPTIMIZED LOADING OF New NETWORKS ON WGS Chris McLain Lino Gonzalez LinQuest Corporation Christopher Swenarton PM WIN-T Fort Monmouth, NJ and ABSTRACT LINK OPTIMIZATION The Wideband Global Satellite (WGS) system represents an order of magnitude increase in capacity over the existing Defense Support Communications Satellite (DSCS) system. In addition to adding capability in the X-band and aKa-band, WGS adds a unique digital channelizer. The digital channelizer greatly increases the flexibility of the WGS payload over previous satellite systems. It allows an individual 125 MHz transponder to be decomposed into as many as 48 sub-channels, each of which can be given its own transfer gain, be cross-banded between X-band and Ka-band, be combined with other sub-channels (fannedin), or be duplicated in multiple downlinks (fanned-out). However, with increased flexibility comes the following planning challenge. Given all of the possible options, how does an operator configure the WGS satellite to most efficiently serve a given network? The principle difference between WGS and a conventional transponded satellite is its digital channelizer. The digital channelizer allows a 125 MHz transponder to be decomposed into 48 sub-channels of 2.6 MHz each. Each subchannel can be assigned its own gain state so that, in principle, individual links can be optimized. This provides a great deal of flexibility over a conventional transponded satellite, which usually only have one, rarely adjusted gain setting per transponder which must be shared by all of the carriers in the transponder. With more flexibility comes more difficulty in optimizing the link setup. The objective of this section is to derive an algorithm for selecting optimum gain states for a specific link. This planning challenge becomes particularly acute for the Network Centric Waveform (NCW). NCW is an advance mesh waveform used by programs such as WIN-T and Prophet. The NCW can support networks of heterogeneous terminals ranging from small mobile terminals to fixed regional hub nodes, which span multiple beams, different bands, and many discontinuous sub-channels. NCW can take full advantage of the flexibility of WGS but doing so presents the operator with a dizzying number of options when setting up a network. This paper describes an algorithm for optimally loading NCW links and networks on WGS. INTRODUCTION The purpose of this paper is to derive a method for optimized the loading ofNCW networks on WGS. This analysis has been divided into two parts: I) the optimization of an individual link, and II) the optimization of loading a collection of links that constitute a network. This analysis will be restricted to the case of a network that resides in a single transponder with a co-coverage area uplink and downlink. Networks with multiple coverage areas, cross-banding, and fanned-in or fanned-out bandwidth segments will be considered in a future analysis. Before proposing a specific optimization method, we will first examine the effect of changing the gain state on an example link: a SNE 1 terminal communicating with another SNE. For this analysis, the transmitter is operated at its maximum ERIP 2 , 42 dBW, and the receiver has a G/T of 11.9 dB/K. The terminals are located in a WGS NCA beam where the G/T is 7.7 dB/K and the saturated EIRP is 60.7 dBW. The uplink center frequency is 30.5 GHz and the downlink center frequency is 20.5 GHz. The transfer gain of the satellite is varied between 177.8 dB and 227.8 dB, reflecting a 50 dB range in gain state in the digital channelizer. C/No Ratios The effect of changing the transfer gain on the C/N 0 ratios of an individual link is shown in Figure 1 for the example link. The uplink C/No is shown as a horizontal, solid blue line, which is not affected by transfer gain. The downlink C/No is a sloped, dashed blue line. As transfer gain increases the downlink C/No also increases. The end-to-end C/No is shown as a solid red curve. The behavior of the end-to-end C/No suggest three regions: 1) the downlink limited region (less than about 197 dB) where the end-toend C/No asymptotically approaches the downlink C/No, 2) the uplink limited region (greater than about 217 dB) where the end-to-end C/No asymptotically approaches the 1 2 978-1-4244-5239-2/09/$26.00 ©2009 IEEE 10f7 PM WIN-T Soldier Network Extension (SNE) SOTM product NeW always attempts maximum supportable burst-rate uplink C/No, and 3) a transition between them (between about 197 dB and 217 dB). These regions will figure prominently in our later discussions. 85 r l I 80 I r -, r I I I 75 Uplink CI No I - - Downlink CI No ~-- - o- End to End C/No 70 ~- - ~- F - - _ 65 ~ til :g. 60 o z U 55 I 45 40 I I ~- - - I I I I I I I I I I I I I I I I .> r l I I I I -:-:/ -~ ---:- --~ -T~ --~~ -:- -~ ~ ~I -: - - - ~ - - - I I - ~ - - 'I - - ~ - - I I - - I - - - I I I I I I I ~- Downlink Power I - - I I I I I I I I I I The downlink power required to achieve the data rates above is shown in Figure 3 for the example link. The solid blue line shows the signal EIRP, which is linear with transfer gain . There is no transition in behavior between regions , so the signal EIRP keeps increasing linearly in the uplink limited region even though the end-to-end C/No is no longer increasing and, more importantly, the data rate is no longer increasing. This is an undesirable condition since downlink EIRP is a limited resource. ~--~---~--~---~--~- ~--~---~--~- Y VI ~--~---~--~---~-- c___-"yJ___ __ __ __ I I I I I I I I ---I---~- I I I ~ I I I ~ I ~ I I - /Dow n Link- ' ~ . ; Limited : L 180 185 190 I I I - - ~ - I I - "-- "--- "-- 0- ~ I -:I ~ - - ~ - - - ~ - - ~/ ~ - - ~ - I 50 I I -- I I· l 195 J ___ ~ __ J_ I I I _L __ ...J ___ I___ .J_ Transition I I I I 200 205 210 I I 215 -1----+-1---4I I I I -1----1.I I I I _L __ J_ I I Lin,kLimited - ~P 220 supported data rate, 32 kbps, requires a transfer gain of at least 193 dB. Data rate increases as a step function with transfer gain until it reaches a maximum of 768 kbps, which is still well below the capability of the modem. After this point, no increase in transfer gain will result in an increase in data rate. 225 55 Transfer Gain (dB) Figure 1. SNE to SNE link C/No versus transfer gain. 00 ~ Figure 2 shows the data rate as a function of the transfer gain for the example link. The solid blue line shows the data rate for a 4.6 dB Eb/No without quantization. The data rate increases linearly with the transfer gain in the downlink limited region . The linear relationship breaks down in the transition region where the data rate increases at a lower rate than the transfer gain, as would be expected from the end-to-end C/No curve in Figure 1. By the onset of the uplink limited region, there is essentially no increase in data rate with increasing transfer gain; the data rate asymptotically approaches a maximum. ==:;=-- -- _ rr--'----'------'----'-------,- - -, - - - ,- - - D:lta Rale : - - Quantized eata Rolle - -.! - - :- - - ~ r--~---r--~---r-- I I I I I ~--~---~--~---~ I I I I I I I I I I I --r--lI I 1---~--1- I I I I [::~:::C::J~--~---I---~ s 8 1d ~---I---~-- - ~--~---~-- I I :C::J:::C::J: -I---~---I----+­ -~--~---~--~-~--~---~--~- ~---:---~--~- -~--~---~--~­ ~--~--- -~--~---~--~- 1 L __ 1 1 ~ __ L --~__ 1 I ~ __ L 1 __ ~ 1 I L __ ~_ c==~=_=c==~= =c==~===c==~= ~--~ --~--~- 1-- --r--~- r--- 1---1---1- ~--~---~--~- ~---~Unk~T7---I---r--~- I 100 185 Urrited 190 I 195 -~--~---~--~­ -r--,---r--,- -1---'---1---1- ~ - - ~ - - - ~ - - ~- -~--~---r--~ - - ~ - Trimsitioo - ~ 1 1 1 200 205 210 Transfer Gain(dB) 1 2 15 -UpUnk- -urrited 220 L __ 1 1 L __ 1 &30 W 25 00 15 1 1 L __ ~ 1 1 1 1 L __ ~ 1 1 1 1 L __ ~ 1 - - 1 1 ~- 1 - - 1 1 ~ 1 1 1 I I . I 1 1 - 100 190 L __ 1 1 1 1 L __ 1 - - ~1 1 -~1 I --l 185 1 1 ~ -r - l- - - -l Downlink I ~ ~~ Lirrited~17--1 1 l- - L __ ~ :--:--V:--~~--~---~ 1 1 1 ~ I 10 1 1 I I 1 1 "- /' '" ~-_-I_--~-_-I~ L __ ~ 1 1 195 1 - - 1 1 I I 1 I L __ ~ 1 1 I I ~ 1 1 ~J '" /' -" 1 __ L ) '" ~ I~ ~ /' ~ r - ~_ L __ 1 1 1 1 L __ ~ ~ 1 1 1 1 1 1 - _~_ 'I __ I -" '1 1 __ _:_ - ~~ _~ 1 - /' " 1 - - 1 1 1 ~_ 1 1 ~_ 1 _:_ - - ~_ -:--:---:---:---:---:1 1 1 1 1 1 1 1 1 1 1 -~--~---~--~---~--~1 1 1 1 1 1 - ~ - - ~ - ~ ~ :- - - ~ - - -UpLinkI - f- - 1 Transition - -+ - 1 -1- - 1 200 205 210 Transfer Gain (dB) - I '. 1 1 . 1 I 215 220 225 -+ - - . Limited. Figure 3. SNE to SNE downlink EIRP versus transfer gain. The situation is even worse when the total satellite down link EIRP is considered, which is shown in Figure 3 as a red dashed line. Total EIRP is the signal EIRP plus the reradiated uplink noise. The difference between the signal EIRP and total EIRP is negligible so very little noise is being reradiated in the downlink limited region . This changes in the transition region where the total EIRP begins increasing faster than the signal EIRP. The difference approaches a value of lO*loglO(l /(Eb/No) + 1) in dB in the uplink limited region. For an Eb/No of 4.6 dB the difference in Total EIRP and signal EIRP is 1.3 dB. This means that not only is the total EIRP increasing in the uplink limited region without any increased in data rate, but the reradiated noise increases the total EIRP by 35% above the signal EIRP. c==~===c==~=-_c==~===c==~= ~ L 1 1 ~ 1 ~---~--~---~-~~- 1 1 Data Rate V L 1 1 225 Figure 2. SNE to SNE data rate versus transfer gain. Downlink Power Efficiency The solid red line shows the quantized data rate based on the thresholds in MIL-STD-188-EEE. Note that the lowest The disconnect between increasing downlink EIRP and fixed data rate in the uplink limited region leads naturally 20f7 to the concept of downlink power efficiency. For this analysis the power efficiency will be calculated as: eq 1 Powerliff = SignalDataRate/l o Signa/OwpIIHackO.ffIlO Based on this definition, it is possible to calculate the fraction of the transponder total power consumed by a signal as its data rate divided by its power efficiency. Figure 4 shows the downlink power efficiency of the example link as a function of transfer gain. The power efficiency is greatest at minimum transfer gain and remains essentially constant across the downlink limited region . The power required in this region only varies in proportion to the data rate. The power efficiency begins to drop off in the transition region . Data rates continue to increase in this region but the power required increases faster and at an accelerating rate. Power efficiency continues to fall in the uplink limited region as downlink EIRP increases but data rate does not. Obviously, this situation and the uplink limited region should be avoided when optimizing a link. L I I L I_ I L L.. I ~: ' ~ ~ I I I I I I_ 1----1---1----1- _L __ I _ _ ' " I , I -, Dow1link SignalEIRP __ I l_:~ -~--4---~ - Dow1link Total EIRP I I __ 1 1 I 1_ -4---1---4- -'---1---''--1---'---'[:::::::: :::<:::::::::: c:::::::: J:::: =[==]===C~ J===I===1: L I L I_ _ L __ l ..!. _ -r--,---~ 1_ _ ' _ I I I I L I L I_ I I I I I I I I I I I I I I I I I I I I ,---,---r----,r---,---r---,- t===,===t===,: 1- __ --1 1- __ -1_ 1---""1---1---"""1- r---,---r----,- ..... - - - ; - - - ..... ---;- I I I I I I I \ -r--'---r--q 1 I ) I -~--~---~--~ \~ I I I I --1---'I -I---~- I I I -r--,---r--'-7 1---'I I I I I I I I ) ~I I I I I I \ ~ I I =t==1===~==!=:=13_=!= __ __ I_L _~ ~ ~ ~ ~_ =~==~===~==~===:=~~~= -~--~---~--~ -1--- ,- - - -OOWilLUnk - ~ ~ r---,---r--""- Umited L __ I I 180 185 L __ ....J 190 200 205 210 Transfer Gain (dB) 215 220 Given these goals, the optimum will likely fall in the transition region . The maximum data rate possible will occur at transfer gains towards the uplink limited edge of the transition region. The maximum power efficient data rate will occur at transfer gains around the downlink limited edge of the transition region. On occasion, the most favorable data rate may occur in the downlink limited region, but only because NCW does not support a higher data rate. Certainly the optimum points are not likely to ever occur in the uplink limited region because the data rate is no longer increasing in this region and the power efficiency is dropping rapidly. The difference between the downlink and the end-to-end carrier to noise-density ratio, delta-ClNo, can be used as a proxy for trade between power efficiency and data rate . We can define the lower edge of the transition region as occurring at a delta-ClNo of 0.5 dB and the upper end at a delta-ClNo of 10 dB. The delta-ClNo corresponds directly to the loss in power efficiency so a 0.5 dB delta-ClNo is a ~ 10% loss in power efficiency relative to the maximum and a 10 dB delta-ClNo is a ~90% loss in power efficiency. This makes delta-ClNo a good ' knob' for setting the optim ization goal. If the user would like the best data rate without losing power efficiency then the permissible delta-ClNo would be set to 0.5 dB. Alternatively, if the user would like the best possible data rate then the permissible delta-ClNo would be set to 10 dB. A balance between the goals of power efficiency and maximum data rate might be a delta-ClNo of 3 dB. An example case where the delta-ClNo is equal to 3 dB is shown in Figure 5. Note that the delta-ClNo of 3 dB, which is marked with a red dot, occurs where the uplink and downlink ClNo 's are equal. This corresponds to a transfer gain of207.28 dB. I ....J_ 195 power. The optimum gain state will be a balance between maximum data rate and maximum power efficiency. 225 Figure 4. SNE to SNE downlink power efficiency Selecting an Optimum Transfer Gain After looking at the effect of changing the gain state for an example case, we will now return to the question of how to choose an optimum gain state and apply the insights we have gained . Of the possible goals for optimizing a link, two stand out: maximum data rate, or maximum power efficiency. Of these, maximum data rate is the most obvious. Maximizing data rate maximizes the possible throughput for a given terminal. Still, there may be cases where the operating with the maximum data rate is undesirable. This can be seen as the data rate approaches the uplink limited region and the power efficiency plummets. If the transponder loading is power constrained, then operating highest possible data rate will consume a disproportionate amount of The corresponding data rate is 384 kbps as shown on Figure 6. An inset in Figure 6 shows the data rate versus transfer gain in more detail around the 207.28 dB point. There are several points of interest to note within the inset in addition to the 3 dB delta-ClNo. First, the 3 dB deltaClNo point, marked with a red dot, is not the lowest transfer gain that could support 384 kbps. The transfer gain could be lowered to 206 dB, which is marked with a green dot, and still maintain the same data rate. This point would actually be more power efficient than using the 207.28 dB and would be a better choice for transfer gain. The only downside of this point is that the link is right on the edge of dropping to a lower data rate. Any small negative perturbation in the link could cause it to fall back in data rate. A compromise to avoid this would be to add a small amount of margin to minimum transfer gain to im- 3 of? prove the stability of the data-rate in the link. The minimum transfer gain plus a half dB of margin, 206.5 dB, is shown as a blue dot in the inset. This is represents the best choice for an optimum transfer gain when the permissible delta-ClNo is 3 dB. 85 r l I , :1 r -I .r I I I 80 F - - - - Uplink CIN o , - _. Downlink CIN o 75 ~ - - I 70 ilj ::g. 60 ~ - .- ~- - ~I I - - I ~ - - ~- I I : .r l I - ~ I I ~ - - - :- - - I ~- - ~- : 3 dB:delta ,C/No:j; I - -:- I I I -r -- -"lI - - ~ ~ I -~ - - - I ~ - - ~ -- I I I -- --- - ;/1 : , : : I I I I I I I I 50 l- __ -l I-_ I I I 1 /-; I I ~- ~ I I y- ~- 180 185 I , - I ~I Y I I I I I I -1- I I I I I I I I I I I _1- __ -1 I I I I 4. - ~ - - ~ - - I I I -1 _ _ 1- __ ....1. _ _ 1 I I 190 195 - I I Lin,kLimited - ~P L - - -l - - -'- - - -l - ' , 200 Transit ion ' , , , 205 210 215 220 Add the transfer gain margin to the minimum transfer gain to yield the optimum transfer gain. Step 2 could speed up by using a binary search through the range of transfer gains to find the maximum permissible delta-ClNo. _..II- - - ~ - - ~ - - ~ - - _:- - - ~ - - ~ - - ~ -- -:Dow n Link-' . . , , Limited ' 1. Compute the uplink, downlink, end-to-end ClNo, and supportable data rate at the minimum transfer gain. 3. Calculate the minimum transfer gain that will support the optimum link data rate. -1----+- ~A~ -x-: ~ - - ~ - - - :- - - ~ - The steps in the algorithm are: 2. Increase the transfer gain incrementally and repeat Step 1 until the difference between the downlink ClNo and the end-to-end ClNo equals the maximum permissible delta-ClNo. The data rate at this point will be the optimum link data rate. ' / Y I t-----l---f----l---l---/ ~--I---_I- ~ 40 T I ~--~---~--~---~--~- L--~---~--~- 55 45 I - - l - "lI - - - I',- - - "lI - - - I',- -/ ' / End to End CI No I I N 65 ° I to the minimum transfer gain for a given data rate. Optimum Transfer Gains fOr Various Antenna Pairs 225 Transfer Gain (dB) Using an implementation of the proposed algorithm, we can now compute for various antenna types. The link budget assumptions are the same as was described for the example case. The EIRP and the G/T of the antenna types analyzed are shown in Table 13 • Figure 5. SNE to SNE link ClNo versus of transfer gain with the 3 dB delta-ClNo point marked. . tics Tabl e 1 A ntenna T vpes an dCh arac t efts I I I I 3dBdeltaC/No o/r (dB/K) , STT+ SNE poprrCN -~-~-~:--I -~-~ Yi I <J) c- oO e- I I I I I - T- - Q) a; I I I I I I I I I ~~ t I I - - - I" - I" , 10' I Yi~ ~ 2C6 - ~ - Transition - ~ 100 185 190 195 200 I 205 210 Transfer <?ei n (dB) - :DEl -r--"---I---" I I I I 215 I I I I I I I I _I __ 1__ ~ _ .: 2fJl IDe :00 211 -UpUnk- Table 2. Optimum transfer gain for various antenna types -Urrit~d220 EIRP (dBW) 74.3 42 .0 53.2 70.0 Table 2 gives the optimum transfer gain for a delta-ClNo of 10 dB (maximum data rate). The selected data rates for the optimum transfer gains are shown in Table 3 and the corresponding power efficiencies are shown in Table 4. Opt. PIl""'"Elf. , I I I 26 .7 11.9 15.1 32.1 LRH ~ -1--:_~!~~a~ni ~s I a:: 8'" , I Receiver (dB) STT+ SNE PoP/TCN LRH 225 Figure 6. The 3 dB delta-ClNo point and possible optimum transfer gains STT+ 178 .9 200 .6 188.4 178 .9 SNE 180 .8 21S.4 203 .2 185 .1 PoP!TCN 178.9 212 .2 200 .0 181.9 LRH 178 .9 19S.2 183 .0 178 .9 Table 3. Data rate at the optimum transfer gain Top Level Algorithm fOr Optimum Transfer Gain Receiver From the discussion in the last section we can derive an algorithm for selecting an optimum transfer gain. 5TT+ STT+ 3072 .0 SNE 768 .0 3072 .0 PoP!TCN LRH 3072.0 (kbps) This algorithm has two inputs : 1) the maximum permissible delta-ClNo, and 2) the transfer gain margin. The maximum permissible delta-ClNo has a range of 0.5 dB to 10 dB and sets the preference between power efficiency and maximum data rate. The transfer gain margin has a range of 0 to 1.5 dB (approximately the spacing between the data rate increments) this sets how close the system will operate 3 4 of? 5NE 3072.0 768.0 3072 .0 3072.0 PoP/TCN 3072 .0 768 .0 3072 .0 3072 .0 LRH 3072 .0 768 .0 3072 .0 3072 .0 STT+, SNE, poP/TeN and LRH are all PM WIN-T satellite terminal s the number of bandwidth segments also minimizes the setup and management complexity for the network. Table 4. Power efficiency at the optimum transfer gain Receiver (kbps/Xpdr) STI+ STI+ SNE PoP/TCN LRH SNE 7,393,766 1,280,446 6,418,318 7,393,766 286,036 42,400 212,530 284,722 LRH PoP/TCN 532,245 23,238,684 88,585 4,439,778 444,039 22,254,673 594,869 23,238,684 Optimization Constraints and Inputs NETWORK OPTIMIZATION NCW networks on WGS will consist of multiple links. If the terminals in a network are of different types or are located at different satellite antenna gain contours, then the links between them will have a different optimum gain states. NCW can accommodate up to 63 bandwidth segments in a network with each bandwidth segment having a different gain state. Networks with more than 63 unique terminal type / satellite gain combinations will require multiple links to share a gain state, which means that some links will have to be operated off of their optimum gain state. Even with networks smaller than this there will frequently be reasons to consolidate multiple links with different optimum gain states into a single bandwidth segment as a result of finite size of WGS sub-channels. WGS sub-channels are limited to increments of 2.6 MHz. If a particular link has a small or negligible data rate then allocating a 2.6 MHz sub-channel to that link is very bandwidth inefficient. This section will deal with the optimization of a whole network in terms of how many bandwidth segments to create and how to optimize their gain states. Optimization Objectives The three most common optimization objectives are: A key constraint when optimizing the network bandwidth gain states is that no link be assigned to a bandwidth segment with a transfer gain lower than its optimum transfer gain. This assumes that the link optimum transfer gains have been chosen to meet certain link objectives (e.g. maximum data rate) and providing a lower transfer gain could preclude those objectives from being achieved. The implication of this is that at least one bandwidth segment must have a transfer gain equal to the highest transfer gain of all of the links in the network. This is also the case if there is only one bandwidth segment in the network. The optimization is also constrained by the maximum power and bandwidth available. These will generally be set by the maximum transponder power and bandwidth. If the optimization objective is the minimum bandwidth, it must be the minimum bandwidth that can be supported within the maximum power available. The optimization is also constrained by the number of bandwidth segments available. While the modem can support up to 63 bandwidth segments, doing so for a single network would generally be undesirable. Based on experience, most single transponder, co-coverage networks can be optimized with one to five bandwidth segments. As a result we include a user settable maximum number of bandwidth segments. The default value is five segments. Finally, we have observed that as the optimization progresses, the network power will asymptotically approach the ideal minimum power for the network. Rather than continue the optimization past the point of diminishing returns, we include a user settable convergence threshold for terminating the optimization within a certain percent of the ideal power. The default convergence threshold is 10%. 1. Minimum Network Bandwidth - If excess satellite power is available or the amount of bandwidth available is very limited, then the objective of the loading may be to occupy as little satellite bandwidth as possible. 2. Minimum Network Power - Conversely, if the satellite is relatively power constrained, then the objective of loading the network may be to minimize the power required by the network. Calculation ofNetwork Power Calculating the network power and bandwidth is essential to optimizing the network. Network power is the sum of the bandwidth segment powers. The segment power is the sum of the throughputs of each link assigned to the segment divided by the power efficiency of the link at the at the bandwidth segment transfer gain: 3. Minimum Total Resources (Bandwidth and Power) More often, the objective will to minimize the limiting resource - either power or bandwidth - so that the network requires the least overall resources. Since power and bandwidth are tradable to some degree, this will frequently lead to a 'balanced' loading where the percentage of transponder power and bandwidth occupied are similar. n_segments NetPower == LSegPower(i) eq2 i=1 . A major secondary goal is to mmmuze the number of bandwidth segments; fewer and larger bandwidth segments will cope with the variability in traffic on a given segment better than many small bandwidth segments. Minimizing m_links SegPower(z)= L )=1 LinklrataThroughputcjs ------~­ LinkPowerEff(SegGain(j)) The link power efficiency as a function of transfer gain is obtained during the link optimization process as shown in 50f7 Figure 4. Calculation ofNetwork Bandwidth It can be seen from Figure 4 that the power efficiency of a link decreases if the link is assigned to a higher transfer gain than its optimum. It follows from this that the ideal minimum power for a network is achieved when there is one bandwidth segment for every unique optimum link transfer gain in the network and every link is assigned to its optimum transfer gain. The ability to compute the ideal minimum power is very useful when determining the quality of the optimization and when to terminate the process. As with network power, network bandwidth is also calculated as the sum of the bandwidth segment bandwidths. The bandwidth segment bandwidth is also calculated as the sum of the link throughputs divided by an efficiency term, in this case the spectral efficiency of the link. The only additional step is that the segment bandwidths are quantized into 2.6 MHz sub-channels: n eq3 NetBW = ISegBW(i) An example calculation of the ideal minimum network power can be done for a delta-CzNo of 10 dB using the example power efficiency values in Table 4 and a set of example throughputs in Table 5. The optimum link transfer gains from Table 2 are plotted against the cumulative network power as a percent of the transponder power in Figure 7. The total for ideal minimum power amounts to 3.9% of the transponder power in this example. SegB wei) = 2.6MHz * ceil(m Table 5. Example scenario data throughputs Receiv er STI+ (kbps) SNE STI+ SNE PoP/TCN LRH PoP/TCN 929 730 977 579 86 262 1,602 29 LRH 237 459 1,926 547 S21 232 978 624 215 210 _ 205 ell .~ 195 f- 190 ~ 185 180 fh } =1 LinkDataThroughput(j) LinkPowerFJJ(SegGairi.,j)) * 2.6MHz J The quantization of sub-channels becomes important when the number of bandwidth segments becomes large and their individual throughput is small. To achieve the ideal minimum network power of3.9% of the transponder power from the example, it was necessary to have all twelve unique transfer gains in the network their own bandwidth segment. Some of these bandwidth segments carried very little throughput. As a result, the required bandwidth for this case would be 25.0% of the total transponder bandwidth . These cases demonstrate that minimum power requires the maximum bandwidth, and the minimum bandwidth requires the maximum power. 200 CJ -* i=l Minimizing the quantization penalty requires minimizing the number of bandwidth segments. If all of the links in the example network were lumped into a single bandwidth segment, then they would require 8.3% of the bandwidth of WGS transponder. On the other hand, as was pointed out before, this case would require 28.2% of the transponder power. 220 ,-----.-------.---~-~-~--,____-_,_____-___, E ~segmenls 1 Optimization Starting Point I r-----r 175 '------ o -L-Q5 --'---1 ---'---1.5 ----'-2 ----" - L-3 Cumulative Percentage of Transponder Power (%) ~5 - :-'--~5 ---' 4 Figure 7. Minimum cumulative power versus transfer gain This example gives every link its optimum transfer gain. Consider what would happen if all of the links in Figure 7 were placed in a single bandwidth segment with a transfer gain equal to the maximum gain link in the network, 215.4 dB. A substantial fraction of the cumulative power occurs in links with optimum transfer gains between 179 dB and 205 dB. Operating these links at transfer gains that are 10 dB to 25 dB than their optimum transfer gains will substantially reduce their power efficiencies. As a result, the cumulative network power would increase to 28.2% of the transponder power - a substantial penalty. This motivates the use of more than one bandwidth segment. The network optimization is an iterative, numerical process. Choosing the correct starting point for the optimization will have a substantial effect on how long it takes to converge. The two obvious alternatives are to start at the minimum power point (maximum number of segments) or the minimum bandwidth point (minimum number of segments). Given our secondary goal of minimizing the number of total bandwidth segments, starting with the minimum bandwidth point makes the most sense. Top Level Algorithm for Network Optimization The basic algorithm starts with a single bandwidth segment with the transfer gain of the highest optimum link transfer gain. The power and bandwidth are then computed for this case. If the optimization objective is minimum bandwidth and the power constraints for the network are already met, then the optimization stops immediately. Adding segments will 60f7 only increase the occupied bandwidth. Similarly, if the objective is minimum total resources and the percentage of transponder power occupied by the network is already smaller than the percentage of transponder bandwidth, the optimization stops immediately. However, if the objective is to minimize power, or to minimize bandwidth with a constrained power, or to minimize total resources when the power required exceeds the bandwidth required for one bandwidth segment, then it is necessary to add bandwidth segments. comes within the convergence threshold of the ideal network power. Example Optimization Figure 8 shows the results of an example optimization. The power threshold has been set to zero and the maximum number of segments has been set artificially high to show what happens with the optimization is allowed to run to the minimum power point. - - Required Power ~ Bandwidth segments are added by splitting an existing segment according to the following steps : I. g i Re-compute the network bandwidth and power based on the new segment transfer gains . Total Resource , , " MinimumTotal \ ,l Resourc/es 15 ____ .gj 8. 10 ~ 5~ . ~ ~ -, ~ over M inimum Minimum Bandwidth oL....::...=.:..:.:::..:..:..:..:::..::..:.:.......L_ 1 Minimum Power Power < 10% c "' 3. Compute the delta power for each bandwidth segment between the actual power required and the ideal power required. This is a measure of how much power could be recovered from the bandwidth segment by splitting it into smaller segments and assigning them transfer gains that are closer to the optimum link transfer gains . 5. , ~ 2. Compute the ideal power for each bandwidth segment by computing the data throughput for each link in the segment and the power efficiency for each link at its optimum link transfer gain. Split the segment with the highest delta power into two segments with roughly equal ideal powers. The upper half of the old segment will have the transfer gain of the old segment. The lower half will have a transfer gain between that of the old segment and the segment with the next lowest transfer gain. - - Required Bandwidth , a: Compute the actual power for each bandwidth segment based on its assigned transfer gain using the data throughput for each link in the segment and the power efficiency for each link at that transfer gain . 4. - - - - Minimum Power "5 25 , 2 3 4 -'-----l' - -- ' -_ 5 6 7 -'-----'-_ -'----'-_ 8 10 9 11 -' 12 Number of Bandwidth Sea men ts Figure 8. Example network optimization Several points on Figure 8 are notable. First, as expected the minimum bandwidth occurs with only one segment. The required power falls off steeply as the first few additional are added. The minimum total resources occur when there are only two segments. While the required power and minimum power lines appear on top of each other from five segments and higher, the two do not actually converge until twelve segments. For practical purposes however, there is no improvement in the network power after four or five segments. The bandwidth however still climbs as segments are added . This example emphasizes the point that only a few bandwidth segments per network are normally required to minimize the network power. Seeking to further optimize the power only results in wasted bandwidth. This validates constraining the maximum number of segments in the optimization and including a convergence threshold. 6. Iteratively adjust the segment gains upwards or downwards and repeat step 5 until the network power is minimized for the number of segments or an iteration limit is reached. Steps I through 6 are repeated until the optimization objectives are met or the maximum segment constraint is reached. If the objective is minimum bandwidth within a constrained power, then the process continues until the power required is reduced to within the constraint. If the objective is a minimum total resource, then the process continues until the percentage of transponder power required falls below the percentage of transponder bandwidth required. Finally , if the objective is a minimum power, then the process continues until the network power CONCLUSION was has the potential to provide an enormous amount of potential capacity to support NCW networks. However, these resources can only be used efficiently with optimized gain states . This paper has proposed a method for optimizing individual link transfer gains for NCW links and the bandwidth segment transfer gains for NCW networks. 70f 7