Series Input Modular Architecture for Driving Multiple LEDs J. Patterson, R. Zane Colorado Power Electronics Center University of Colorado at Boulder Boulder, CO 80309-0425 USA Email: {j patters, zane}@colorado.edu Abstract – This paper introduces a modular power converter architecture based on series input connected converter cells with independent outputs that each drive a small series string of LEDs. The approach provides the benefits of operating from a high bus voltage while using low voltage cells that independently regulate LED currents for improved light uniformity and LED failure mitigation. Stability of the series converter string and independent regulation of the floating LED output currents are achieved using a two-loop controller that relies on chain control communications between cells to share system information. Experimental verification of input bus voltage sharing and independent cell current regulation is provided using a prototype of the proposed system with four series buckboost converter cells, where each cell drives four 700 mA LEDs. Ig ILED1 + V1 _ Vg Cell1 + – ILEDn Additional Parallel Strings + Vn _ Celln Keywords – solid-state lighting, light emitting diodes, LED, digital control, modular systems, series input, power converter I. INTRODUCTION Significant improvements have been achieved over the past decade in solid state light emitting diodes (LEDs), leading to high efficiency and high brightness solutions that are expected to surpass and replace existing lighting systems in many applications over the coming decade [1,2]. Although the trend is towards development of higher current and brightness LEDs, practical solutions for replacing large lighting panels in general lighting applications still require use of many LEDs in a system [3]. Typical solutions place many if not all of the LEDs in a series string when operating from a high voltage supply such as the rectified ac line. In addition, current limiting and regulation of each LED string is required to achieve the specified luminous output and light uniformity in the presence of variations in the LED forward voltages [4] and applied input voltage. Existing solutions for current regulation in LED strings include use of series resistors or linear regulators [5] or a single high voltage switching converter [6] per string. Each of these solutions has drawbacks, including low efficiency for series linear elements and high component stress and cost for high voltage converters. Existing solutions also have limited flexibility, requiring all LEDs in a large series string to operate at the same current. The limitations make it This work is co-sponsored by the National Science Foundation (under Grant No. 0348772) and industry sponsors of the Colorado Power Electronics Center. 978-1-4244-1668-4/08/$25.00 ©2008 IEEE Figure 1 – Proposed Modular Architecture difficult to compensate for variations in light output among LEDs and individual LED failures in the system. This paper introduces a modular architecture, as shown in Fig. 1, based on series input connected power converter cells with independent outputs that each drive a reduced number of series LEDs. The converter cells share the same input current but self-adjust their respective input voltages to properly share the total input bus voltage. The result is a modular architecture that can operate from a high bus voltage using low voltage converter cells. The approach has the flexibility to independently regulate the currents in smaller sub-strings of LEDs for improved light _ uniformity and LED failure mitigation. II. SYSTEM DESCRIPTION A. System Analysis A block diagram of the system structure is shown in Fig. 1, with detail given for one series string of converter cells. The number of converter cells in a string is based on the voltage rating for each cell. The total number of LEDs associated with a string (sum of LEDs at output of each cell) depends on the size of the system, but is preferably selected to maximize efficiency by setting the nominal converter conversion ratios to near unity. The approach can be extended to larger systems by adding additional parallel strings of converter cells. 2650 Authorized licensed use limited to: UNIVERSITY OF COLORADO. Downloaded on November 12, 2008 at 13:04 from IEEE Xplore. Restrictions apply. The input characteristics of the ith and jth cells in a series string of n cells for the modular architecture are governed by the following relationships Vj = η i Pj , Vi η j Pi ⎛ η n Pj Vg = V1 ⎜1 + 1 ∑ ⎜ P1 j =2 η j ⎝ Ig = ∑ i (1) ⎞, ⎟ ⎟ ⎠ (2) Pi , ηiVg (3) where Ig is the input current, Vg is the input bus voltage, Vi, Vj are cell input voltages, Pi, Pj are cell output powers, and ηi, ηj are cell efficiencies. The converter cells ideally act as constant power sinks in the input series string and transfer regulated power to the output LEDs. This behavior is shown in Fig. 2 with the input I-V characteristic for two cells, while more cells would have similar characteristics. The startup and breakdown regions represent operation below and above the rated voltages of the converter cells, Vmin and Vmax, respectively. Unfortunately, this type of series converter operation is prone to instability, as shown in similar architectures [79]. At least one of the converter cells will always reach steady state outside of the desired operating region, which makes stabilization of the system the primary concern. The main control objective, while ensuring stability of the system, is to have proportional input voltage sharing according to (1) while maintaining output current regulation of the series LEDs. Two significant differences in the LED system application compared to previous series architectures are that (1) the converter cell outputs independently drive floating LED loads and are thus not connected in series or parallel and (2) the output power or LED current regulation loop in each cell can be slow since Ii Constant Power Region Desired Operating Points Ig Cell 1 Cell 2 Startup Breakdown Vmin Vmax Figure 2 – Input I-V characteristics for 2 cells operating as constant power sinks Vi it only has to respond to gradual variations due to temperature and aging. Based on the characteristics mentioned above, there are two proposed control approaches: an autonomous open loop approach and a closed loop approach using inter-cell communication. The open loop approach requires no inter-cell communications, has a simple control implementation and a preset nominal operating mode. The closed loop solution achieves both stable input voltage sharing and independent output current regulation. The approach does require inter-cell communications, but maintains modularity with identical cells and communications are only preformed between neighboring cells to avoid isolation requirements. B. Open Loop Control Approach The open loop solution is based on having each converter cell behave as a constant loss-free resistor (LFR) [10] at its input port. This approach provides stability to the series converter string and ensures that each cell shares the input bus voltage proportional to its relative power in the system. The approach is termed “open loop” since the output current of each cell is not regulated. The converter cell may or may not have an internal feedback loop to achieve LFR behavior, depending on the converter topology and operating mode. One simple solution is to use a buck-boost converter operated in the discontinuous conduction mode (DCM) [10]. In this case, the averaged input resistance Re of the cell is a function of the duty cycle D as given by Re = 2L , D 2Ts (4) where L is the inductance and Ts is the switching period. Given nominal operating conditions, the cell duty cycles would be preset at the factory or calibrated by the user to generate the desired nominal LED output currents based on a specified number of cells and nominal input voltage Vg. The approach is simple and has several added benefits including the ability to dim the LEDs by scaling Vg and inherent power factor correction (PFC) due to the resistor emulation of the input port. Drawbacks include the requirement for preset configuration and the inability to react to system variations such as changes in Vg, component values or LED failures. C. Closed Loop Control Approach The closed loop approach requires the converter cells to operate with two control loops in a similar fashion to PFC converters [10], as shown in Fig. 3. A fast inner loop, similar to the open loop approach, is used to achieve resistor emulation at the converter cell input port. The inner loop again provides stability to the series converter string and ensures proportional bus voltage sharing. A second slow outer loop with an integral type compensator is used to regulate the cell’s LED current or power. This approach ensures that the converter cells stabilize to the desired input voltage while remaining robust to changes in Vg and regulating the LED current for improved light uniformity and LED failure response. Since there are multiple competing loops within the system, it is necessary to consider system stability. If DCM buck-boost cells are again used to achieve near LFR 2651 Authorized licensed use limited to: UNIVERSITY OF COLORADO. Downloaded on November 12, 2008 at 13:04 from IEEE Xplore. Restrictions apply. behavior, then the desired duty cycle of a single series cell can be written as Ig n D1 = Additional Converter Cells with Compensation and Chain Control Rg ∑η P 2 L i =1 i i , ⋅ 2 η1 P1 Vg Ts (5) Vg + – where Pi is the cell output power. Given a nominal input bus voltage Vg and assuming near identical cells with efficiencyη = η i , the duty cycle can be approximated as P D1 ≈ K tot , P1 I1 P1 + V1 _ R1 Duty Cycle Generation (6) D1 Ptot_next Chain Control where the constant K is given by K= (7) and the total output power Ptot is given by n Ptot = ∑ Pi . (8) i =1 In order to properly control the system, it is necessary to look at the potential change in duty cycle given the dynamics of the cell output power. This sensitivity function is given by ∂D1 =K ∂P1 Ptot 2 , (P1 ) (9) 3 2 which shows that the gain in the outer feedback loop will have a strong dependence on the total power Ptot as well as the cell’s own power Pi. In fact the polarity of the sensitivity function changes when the power of a cell is equal to the total power of all other cells. For this reason, the knowledge of Ptot is required to ensure the duty cycle is properly set. ILED1 Ptot_prev Figure 3 – Resistor emulator cell with closed loop controller within distributed architecture The communication approach taken involves passing Ptot among the cells in order for each cell to update its duty cycle corresponding to the received information. To maintain modularity in the system and avoid the cost and complexity of a system controller, a digital moving window average or chain control is adopted [11]. Each cell transfers in sequence the moving average of Ptot to the next cell in the chain via a shared node according to Ptot _ next [n] = Ptot _ prev [n] − Pled _ i [n − 1] + Pled _ i [n] . (10) Figure 3 shows the LFR converter model with the closed loop control approach. The controller consists of three blocks, with a more detailed view shown in Fig. 4. The current control block contains an integral compensator which takes the measured error between a reference and the output current and predicts the next desired output current. The output power can be approximated as a linear function of the output current when the converter cell is not in the startup or breakdown regions, therefore in closed loop the predicted output current is proportional to the predicted output power of the cell. Ptot_prev [n] ILED [n] ILED_ref Current Control Controller 2 Lη , 2 Vg Ts P1 − + VLED1 _ Pled_i [n] G Ptot_next [n] Di [n] z -1 LUT Clookup [n] Pled_i [n-1] Current Control Chain Control Duty Cycle Generation Figure 4 – Closed loop control block diagram as implemented with DCM buck-boost cell 2652 Authorized licensed use limited to: UNIVERSITY OF COLORADO. Downloaded on November 12, 2008 at 13:04 from IEEE Xplore. Restrictions apply. This predicted output power Pled_i[n] is then fed into the chain control block. A new total power Ptot_next[n] is calculated using the passed in total power Ptot_prev[n] from the previous cell and the previous and current value for the predicted cell output power as given in (10). The new total power is then passed to the next cell in the chain. The duty cycle generation block takes Pled_i[n] and Ptot_next[n] from the chain control block and calculates the cell’s duty cycle from (6) using only one multiplication Di [n] = Clookup [n] ⋅ Ptot _ next [ n] , (11) and a lookup table (LUT) to find the constant Clookup[n] from its input Pled_i[n] according to Clookup[n] = K . (12) Pled _ i [n] In order to implement amplitude dimming with the closed loop algorithm, K would need to be scaled with the input voltage Vg which could be implemented with a LUT in order to avoid the need for sensing Vg. One simple way to implement LED failure mitigation would involve sensing a fault condition such as excessive average current or zero current. The converter cell could then short its input allowing no power to be transferred to the load while preserving continuity of the input current through the rest of the series string. The controller ensures that the remaining cells redistribute the input voltage while maintaining current regulation. III. SIMULATION A. Simulation Setup For the modular architecture, the converter cell with resistor emulation was modeled in Simulink. Using the LFR model shown in Fig. 3 and a basic exponential model Ptotprev Ig T1 cell_1 Ptotprev Ig v1 v2 v3 v4 Ig T2 cell_2 Ig Vin Ptot cell_3 Ptotprev Ig T4 Vin Ptot Ptotprev T3 Vin Ptot Vin Ptot cell_4 Figure 5– MATLAB Simulink top level simulation with offset sampling from (10) indicated for the LED load, both control approaches were simulated. Because the chain control relies on sequential communication between cells, a discrete sampling algorithm was implemented. Using a communication sample time T1 for the bottom cell, three offsets were created at T2 = T1 + T1 , 4 T3 = T1 + T1 , and 2 T4 = T1 + 3T1 4 (13) for each of the remaining three cells. There was also a specific startup procedure in the chain control to ensure a proper simulation. In this manner, each floating cell has ideal communication keeping the moving window accurate. Figure 5 shows the top level Simulink model implementation of chain control where the converter and compensator model of Fig. 3 is inside each cell block and input current Ig is generated from the voltage across a source resistance Rg. B. Open Loop Results A simulation using the open loop control approach was performed to validate resistor emulation techniques and show the effects of dimming. The open loop simulation was implemented by ignoring the chain control information and setting the duty cycle inside the model to a fixed value. Four resistor emulator operated converter cells were series connected with fixed duty cycles. Amplitude dimming was then simulated by initializing a bus voltage at Vg = 50 V and then performing a step transition to Vg = 40 V. The simulations verified the four cells acted as resistors sharing the bus voltage and consequently decreasing output current appropriately when the bus voltage was reduced. C. Closed Loop Results Closed loop control of the system was also simulated with several goals: to verify stable input voltage sharing while maintaining output current regulation and to validate the ease of LED failure mitigation. Four series connected cells were again simulated using the resistor emulator model but now with the chain control and integral compensator enabled. Figure 6(a) shows all four cells initially in regulation with 620 mA output current. Two successive transients were applied to cells 4 and 2 respectively. Both transients were applied via an LED reference current change to 402 mA. After applying the transient in both cases, the duty cycle of the applied cell increased to obtain the new regulated current and correspondingly, shown in Fig. 6(b), its emulated resistance decreased causing its share of the input voltage to decrease. Simultaneously, the other three cells’ input resistances increased, effectively regulating to the same output current. A second simulation was performed to show LED failure mitigation. Assuming the cell could detect fault conditions (i.e. excessive currents or open circuit) and physically short its input, the simulation was created to show the resultant dynamics of the remaining cells. The transient in Fig. 7 shows the remaining three cells redistributing the input voltage and regulating their output currents. Table 1 gives the relevant DC operating points. 2653 Authorized licensed use limited to: UNIVERSITY OF COLORADO. Downloaded on November 12, 2008 at 13:04 from IEEE Xplore. Restrictions apply. (a) (a) (b) (b) Figure 7 – LED failure and input shorting of cell #1 (a) output current (b) input I-V trajectory Figure 6 – Closed loop transients on cell #4 and #2 respectively (a) output current (b) input I-V trajectory TABLE I - DC OPERATING POINTS FOR CLOSED LOOP SIMULATIONS Bus Input V g [V] C.L. Transients Fig. 6 LED/Cell Failure Fig. 7 Initial State Final State Initial State Final State IV. 50 50 40 40 I g [A] 1.045 0.865 1.051 0.833 Cell Input Voltage Cell Output Current V 1 [V] V 2 [V] V 3 [V] V 4 [V] 12.42 15.13 8.36 0.22 12.42 9.75 8.36 10.54 EXPERIMENT A. Experimental Setup A prototype was developed to experimentally verify the modular architecture. The circuit employs the modularity of the architecture by having separable converter and load stages. Each cell is controlled via an onboard floating FPGA and corresponding sensing circuitry and ADCs. In order to power the FPGAs and other ICs reliably, two auxiliary DC-DC converters are driven by the cell input voltage with a UVLO that prevents operation below 6.3V on any cell. Given the power sink nature of the auxiliary 12.42 15.13 10.44 13.18 12.42 9.75 12.53 15.81 I led 1 [A] 0.620 0.620 0.450 0.010 I led 2 [A] 0.620 0.402 0.450 0.450 I led 3 [A] 0.620 0.620 0.450 0.450 I led 4 [A] 0.620 0.402 0.450 0.450 converters, maintaining stability during startup is a challenge. A 100 Ω resistor was placed across the cell input to stabilize the voltages during startup when the FPGAs were connecting to a CPU which was used for virtual I/O and logic analyzer functions only. Once the series string of converters reached a steady state open loop condition, their emulated resistances were much less than the 100 Ω resistor negating its effect. A four switch buck-boost power stage, as shown in Fig. 8, was implemented to test all three basic topologies (buck, boost, and buck-boost). A DCM buck-boost configuration was chosen in order to provide simple resistor emulation for the inner control loop, although 2654 Authorized licensed use limited to: UNIVERSITY OF COLORADO. Downloaded on November 12, 2008 at 13:04 from IEEE Xplore. Restrictions apply. Ig ILED2 L2 I2 + + FPGA V2 Cin2 Cout2 _ Isolation VLED _ Ig ILED1 L1 I1 + + FPGA modular load configuration to test a variety of loads. A four LED load was chosen for each cell and the experiments covered an LED current range of 300 mA – 800 mA. Each cell controller was implemented digitally in the floating FPGA, while an isolation IC was used to communicate between cells, validating the function of the system without a central global controller. A differential sensing circuit and an 8-bit ADC was used to convert the sensed output current. The main digital hardware for the controller itself was one LUT and one multiplier. A delay line based hybrid DPWM was used for the gate drive signal [15]. Figure 8 shows the prototype design and Table 2 gives details of the hardware used. V1 Cin1 Cout1 _ VLED _ Figure 8 – Prototype design using isolated FPGAs with Li = 1 μ H, Cin_i =1 μF. C out_i = 22 μ F average current [12], hysteretic [13] or nonlinear carrier control [14] methods could be used with a variety of topologies and operating modes to achieve the same result. Luxeon K2 700 mA emitters were used with a B. Experimental Results Many experiments were performed with this prototype using different topologies, operating modes and conditions. However, the two experiments shown in Figs. 9-10 validate the key aspects of the two proposed control approaches for this architecture. First, all four cells were run using the open loop control approach and DCM buck-boost power stages with four LEDs each. All four cells were initialized at a duty cycle that yielded approximately 620 mA output current given a bus voltage Vg = 50 V. Then amplitude dimming was tested by quickly decreasing the bus voltage to Vg = 40 V. Figure 9 shows that the four cells respond identically by decreasing their output current according to the decrease in input voltage. A second experiment is shown in Fig. 10 which tested the closed loop control approach with two successive Figure 9 – Open loop control showing uniform dimming Ch-1: I LED1, 500 mA/div; Ch-2: I LED2, 500 mA/div; Ch-3: I LED3, 500 mA/div; Ch-4: I LED4, 500 mA/div; Time: 1 s/div Figure 10 – Closed loop transients on cell #4 and #2 respectively Ch-1: I LED1, 500 mA/div; Ch-2: I LED2 , 500 mA/div; Ch-3: I LED3, 500 mA/div; Ch-4: I LED4, 500 mA/div; Time: 1 s/div TABLE II - POWER SWITCHES AND IC COMPONENTS USED IN PROTOTYPE Device Part Number Description Diode PDS540 Schottky 40V 5A MOSFET STN3NF06L N-Channel 60V 4A ADC AD7825 8bit 4Channel Multiplexed HB Driver LM5101 Half Bridge Driver Opamp OPA350U 35MHz Rail-to-Rail Isolator ISO7221 Dual Digital Isolator 2MSPS Boost LT3436 Regulator 3A Buck LT3508 Dual Regulator 1.4A TABLE III - DC OPERATING POINTS FOR EXPERIMENTS Bus Input Cell Input Voltage Cell Output Current V g [V] I g [A] V 1 [V] V 2 [V] V 3 [V] V 4 [V] O.L. Dimming Fig. 9 C.L. Transients Fig. 10 Initial State Final State Initial State Final State 50 40 50 50 1.079 0.876 1.040 0.885 12.62 9.83 12.60 15.77 12.33 9.96 12.45 9.26 12.29 10.14 12.46 15.63 12.89 10.15 12.92 9.80 I led 1 [A] 0.625 0.373 0.635 0.640 I led 2 [A] 0.618 0.353 0.625 0.398 2655 Authorized licensed use limited to: UNIVERSITY OF COLORADO. Downloaded on November 12, 2008 at 13:04 from IEEE Xplore. Restrictions apply. I led 3 [A] 0.646 0.401 0.610 0.612 I led 4 [A] 0.626 0.388 0.620 0.402 small transients. The experimental results were obtained using the same conditions as the simulation from Fig. 6 and correspond well to the simulation results. When a transient change in the reference current occurred on one cell, it changed its duty cycle to regulate to the new value, while the remaining cells were able to maintain their current regulation with only a small perturbation. Table 3 contains the DC operating points of the experiment for both the open loop dimming test and the closed loop transient. The experimental results for the two consecutive transients closely match the simulation of the closed loop chain control architecture. C. Observations The experimental prototype was designed to be flexible and stable through many different testing conditions. The FPGAs were chosen to make control algorithm changes quick and to have the ability to synchronize with a CPU via JTAG communication for virtual I/O and logic analyzer capabilities. The JTAG communication is in a daisy chain structure and an isolation circuit was needed to connect all the FPGAs to the CPU. This communication with the CPU was not used to stabilize the system, but rather to provide simple monitoring and triggering of operating modes. The auxiliary converters used to power the FPGA, ADC, op-amp and drivers were also used to simplify the experimental process. A potential commercial application is to integrate the control functions and power devices on a single low voltage IC. The only external components are the inductor and output capacitor, which can be minimized with high frequency operation of the low voltage cell. Each integrated cell would need only two control pins, one for sending signals up the chain and one for sending signals down the chain as well as the power stage pins. Isolators are not required for communications since control pins only connect to neighboring cells where current-mode techniques could be used to communicate over these overlapping nodes. A diagram of one option using a buck-boost converter power stage is shown in Fig. 11. The benefit when compared to one large high voltage converter is that many miniature, high efficiency, reduced voltage and power cells can be operated directly off of a high voltage bus and grouped with sub-sets of LEDs and distributed in the lighting fixture. V. CONCLUSION The proposed series input modular architecture for LED lighting systems achieves the benefits of operating from a high bus voltage using low voltage converter cells with the flexibility to independently regulate the currents in small sub-strings of LEDs for improved light uniformity and LED failure mitigation. Stability of the series converter string and independent regulation of the floating LED output currents are achieved using a two-loop controller that relies on chain control communications between cells to share system information. Simulation and experimental results demonstrate feasibility and operation of the architecture using series buck-boost converter cells. Both dimming and LED failure mitigation possibilities are enumerated and briefly examined as added benefits of the architecture. Additional Converter Cells Ig Vg + – + Vi _ Comup Integrated Control and Power Switches ILED_i Li Cout_i + VLED_i _ Comdown Additional Converter Cells Figure 11– Proposed integrated converter cell REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] M. Krames, O. Shchekin, R. Mueller-Mach, G. Mueller, L. Zhou, G. Harbers, and M. Craford, “Status and future of high-power light-emitting diodes for solid-state lighting,” Journal of Display Technology, vol. 3, pp. 160-175, 2007. M. Shur and R. Zukauskas, “Solid-state lighting: toward superior illumination," Proceedings of the IEEE, vol. 93, pp. 1691-1703, 2005. J. W. Curran and S. P. Keeney, “Replacement of fluorescent lamps with high-brightness LEDs in a bridge lighting application,” in Proc. Sixth International Conference on Solid State Lighting, vol. 6337, no. 1, pp. 633-719, 2006. LUXEON Product Binning and Labeling, Application Brief AB21, Philips Lumileds Lighting Company, May 2006, [Online], Available:URL: http://www.lumileds.com/pdfs/AB21.pdf Datasheet: TLC5940, “16 Channel LED Driver w/DOT Correction & Grayscale PWM Control,” Texas Instruments Incorporated, [Online], Available: http://focus.ti.com/docs/prod/folders/print/tlc5940.html Datasheet:HV9910, “HV9910 Universal High Brightness LED Driver,” Supertex, [Online], Available: www.supertex.com/pdf/datasheets/HV9910.pdf K. Siri, M. Willhoff, K. Conner, “Uniform Voltage Distribution Control for Series-Input, Parallel-Output Connected Converters,” in Proc. IEEE Aerosp. Conf., Mar. 2005, vol.2, pp. 1-9. P. Krein, J. Kimball, “Series-parallel approaches and clamp methods for extreme dynamic response with advanced digital loads, “ IEEE PESC 2005. pp. 1441-1445 A. Binge, N. Mohan, R. Giri. R. Ayynar, "Semi-parallel connection of dc-dc converter modules with active sharing of input voltage and load currents" in Proc. IEEE APEC, 2002, pp. 648-653. R. Erickson, D. Maksimovic, Fundamentals of Power Electronics, Second Edition, Copyright 2001 Y. Zhang, R. Zane, D. Maksimović, “Current sharing in digitally controlled masterless multiphase DC-DC converters,” IEEE PESC 2005. J. Rajagopalan, F. C. Lee, P. Nora, “A generalized technique for derivation of average current mode control laws for power factor correction without input voltage sensing,” in Proc. IEEE APEC, 1997, vol. 1, pp. 81-87. J. Spangler, A. Behera, “A comparison between hysteretic and fixed frequency boost converters used for power factor correction,” in Proc. IEEE APEC, 1993, pp. 281-286. R. Zane, D. Maksimović, “Nonlinear-carrier control for High Power Factor Rectifiers based on Flyback, Cuk, or Sepic Converters,” in Proc. IEEE APEC, 1996, pp. 814-820. V. Yousefzadeh, T. Takayama, D. Maksimović, “Hybrid DPWM with Digital Delay-Locked Loop,” in Proc. IEEE COMPEL, 2006, pp. 142-148. 2656 Authorized licensed use limited to: UNIVERSITY OF COLORADO. Downloaded on November 12, 2008 at 13:04 from IEEE Xplore. Restrictions apply.