High-power high-fidelity switching amplifier driving ... - IEEE Xplore

77 downloads 92 Views 459KB Size Report
General Electric-Global Research Center. Qiming Li. GE-Global Research Center. William F. Wirth. General Electric Health Care. Niskavuna. NY 12309. U.S.A..
2004 35th Annual IEEE Power Eleclronics Specialisrs Conference

Aachen, Germany, 2W4

High-Power High-Fidelity Switching Amplifier Driving Gradient Coils for MRI Systems

Juan Sabate, Luis .I. Garces, Paul M. Szczesny General Electric-Global Research Center Niskavuna. NY 12309. U.S.A. Email: [email protected]

. ,

Qiming Li GE-Global Research Center Shanhai. 200233 -China

Abstract- Magnetic Resonance Imaging (MRI) systems use an amplifier to drive the gradient coils with several hundred amperes and voltages i n excess of 1500 V. The fast response and extremely low coil current errors needed for good image quality i n M R I systems are one of the main challenges for the designer. This paper presents new power stage architecture for the gradient amplifier, and two different control modulation techniques. The power stage consists of three full-bridges i n a stack configuration operated with interleaved switching to deliver the required output voltage with high-frequency current ripple. One of the bridges has 400V input voltage and switches at high frequency to provide high bandwidth, the other two bridges have 8OOV and switch at lower frequency for loss reduction. The modulation technique i s crucial t o achieve the required waveform accuracy; a fully digital control provides great flexibility on the control. The paper explores two modulation options; one that minimizes the switching losses, but that will require hi-directional power supplies for some waveforms, and another with higher losses but more suited for arbitrary waveforms. The experimental results show the waveform accuracy from the amplifier implementation with the selected control.

I /

I I

Fig. 1. Amplifier with two full bridges and a linear stage

1. INTRODUCTION The gradient amplifiers in Magnetic Resonance Imaging (MRI) drive the gradient coils with currents in excess of several hundred amperes to create the gradient fields used in the imaging process. The gradient fields have to be modified at frequencies of up to a few kHz for fast imaging, and for the typical inductances of the coils on the range of several hundred uH to a mH, the voltages required are in excess of ISOOV. The image quality and resolution depend greatly on how precisely controlled are the applied fields. High performance systems require high accuracies on the currents to prevent image artifacts. The high power and high current accuracy needed pose great challenges to the designer’s selection of the power stage and the control. The solutions repotted involve always structures with bridges in parallel [2,3,6] or bridges stacked [1,7-9,1 I ] to be able to meet the requirements with the current existing semiconductor devices. Earlier solutions used linear amplifiers for their high fidelity, but they became impractical for the power levels required nowadays. This lead to hybrid solutions combining a linear amplifier, to provide the bandwidth and accurate control, with switched power stages to boost the voltage for the fast transitions [IO]. Fig. 1 shows a simplified schematic for a

0-7803-8399-0/04/$20.0002004 IEEE.

William F. Wirth General Electric Health Care Waukehsha. WI 53 186 USA

I

Fig. 2. Proposed amplifier power stage

hybrid amplifier. The linear amplifier stage becomes totally impractical for the high current levels of new systems. The solution proposed in this paper, Fig.2, consists on replacing the linear amplifier by a fast switching bridge (62.5kHz),

26 1

Aochen. Germany, 2004

2004 3Slh Annual IEEE Power Electronics Specialists Conference

and two other bridges switching at lower frequency (31.25kHz) to provide the higher voltage. The amplifier can be implemented with lower voltage ratings fast IGBTs on the high frequency bridge, and higher voltage rating but slower IGBTs for the low frequency bridges. The interleaved operation of the three bridges provides a high frequency low amplitude ripple. Two different PWM control schemes are presented in this paper for the proposed converter. One that has the highest ripple frequency and can be applied to arbitraly waveforms without limitation, and another that minimizes the losses in the amplifier, but that requires a more complicated control, and is not suited for some specific type of waveforms. The fully digitally control used in the implementation can be used for both options, but more importantly has allowed to introduce compensations for all the system non-idealities, like dead time or input voltage variation. The comparison of the different control options is based in a simulation model, and the preferred option has been already implemented in hardware and the initial results shown confirm the predicted operation. """.."_.~~^_~~x"x.xxx""xxx"x"xIx

Control Stage

11.

AMPLIFIER DESCRIPTION

Three inverter bridges are connected with their outputs in series to supply the high voltage, high current required by the gradient coil, which is almost an ideal inductor. Two bridges are supplied with higher voltage, switching at lower frequency, 31.25kHz. The other bridge is supplied with lower voltage, switching at higher frequency, 62.5kHz, with faster switching devices. In this way, both the fast response, thanks to the low voltage bridge, and the high voltage capability, driving by the two high voltage bridges, could be achieved simultaneously. The power supplies to generate the input voltages to the bridges are not bi-directional. Since the load is passive there is no net energy returned to the source. However, because several voltage sources are available, special care has to be taken to select a modulation that does not result in charging the coil field with one supply and discharging it into a different one. This would result in a voltage increase at one power supply output from energy delivered from one of the other outputs. This over voltage could not be regulated bi the unidirectional power supply, and it can exceed component ratings an cause excessive additional losses. There are several modulation techniques to control this converter. Option I In this control option the two higher voltage bridges, shown in Fig.3, provide the inductance voltage drop (LdVdt), and the low voltage bridge provides the resistance voltage drop in the coil and the feedback. Three bridges switching is interleaved to achieve high output ripple frequency and thus reduce the coil current ripple. Fig. 4 shows the typical waveforms for this option when the amplifier drives the gradient coil with a sinusoidal current. The modulation used in this option is shown in Fig. 5 The high voltage and the low voltage bridges have different commands: feedfonvard for the high-voltage ones and

A.

i I,

Fig. 3 Block diagram for Option I 1w c

v

Fig. 4

Option I typical waveforms: Coil voltage (top), lerror (middle), lreference (bottom)

262

-

?

,

,

,

,

,

,

I

,

I

,

I

,

Aachen. Germany, 2004

2004 35rh Annual IEEE Power Elecrronics Speciolirrs Conference

feedback for the low voltage one, Fig. 3. Figure 5 shows an example in which the low voltage bridge has a duty cycle, D, less than 0.5 and the high voltage bridges have a D larger than 0.5. In this option the high voltage bridges provide .the reactive power delivered to the coil, while the low voltage bridge delivers the coil losses and corrections to the feedforward. In the practical circuit the feedforward command will not be exactly L(dddt), and some amount of reactive energy could be pumped into the low voltage supply. However, the feed forward error will be small and the excess energy can be easily dissipated in the coil and bridge losses. Operation with this control option ensures that no over voltage will occur for any kind of coil current waveform. B.

TABLE I INITIAL CONTROL LOGIC FOR OPTION 2

TABLE II MODIFIEOCONTROL LOGIC FOR OPTION 2

Option 2

The goal of this option was to minimize the switching losses in the amplifier. The voltages selected for this option were 8OOV and 400V for the high voltage bridges and low voltage bridge respectively. The control for the three bridges uses a command calculated with the feed forward and the feedback, as shown in Fig.6. The bridges used to deliver the required voltage are combined as shown in Table I control logic. In order to achieve the lowest switching losses, the two high voltage bridges, supplied with 800V are not pulse width modulated, and only provide voltage when more than 400V are needed; otherwise they remain in freewheeling mode with no output voltage. According to the voltage range required 400V1200V, or 12OOV-2OOOV, one or the two SOOV bridges are set to add voltage. The low voltage bridge, supplied with 400V, is the only one switched all the time at 62.5kHz, regulating the required output voltage. Fig.7 shows the _.. ...................... .................... .."~ ~

CantmlStage

;/

and output voltage for each bridge, the output voltage is scaled as lV=200V. However, because the input voltages to the amplifier are provided by a power supply that cannot return power to the grid, the power delivered has to be always positive or zero for each of the voltage supplies. The control presented for this option has an energy balance problem for some specific waveforms. For example, for the current waveform shown

Powerstam* "

.

~w Ldildt

........ Fix. 8 Example ofenerxv balance problem

:...... I".., , ........... .................. Fig. 6 Block diagram for Option 2

::'..........................

l.. .....i

N

Fig. 9 Simulation results for EPI waveform with l200V drop on the inductance

Fig. 7 Output voltage versus control voltage

263

Aochen, Germany, 2004

2004 35rh Annual IEEE Power Elecrronics Specialists Conference

.. .. ,

/

,

WMCq.WJ

I I

,

,

,

I.,:nli

WqM-t"J

. .

5

..

nli

Fig. 10 Output voltage versus control voltage

in Fig.7, if L,,l*(d~dt)=1200V, and because the resistive voltage drop on the inductor, the voltage during interval TI is in the range 800V to 12OOV, while during interval T2 it is within the range from 12OOV to 16OOV. As a result, the coil is charged with energy from the 800V power supplies and it is always discharged into 400V, which would result in the 400V supply voltage increasing. The simulation results for this case, Fig.9, shows the voltage increase on the 400V SUDDh'.

c-

The modification ensures that energy pumped into 400V

..

.., out from 400V. Fig.11 shows the ""$pp&e&Y2* _ ,, c . &, ,~ ~, . i ,~ ~ ~ ~ v , , 4equals 0 0to vthat, pumped , -. , simulation results of the modified control under the same ! 1 .~ ' : '4 ;:' 0 -::_; o v condition of Fig. 11,p_.~. ;~ - 8. The average - input . currents for the three

-0 E....

*j

I

I~

~

1

I

I.

I

~

!''

~~

I*

I.+A&&&-&+ , &$". r ~ - - --.:+~ .

-

--."-&.%

I

a . ..L.

Fig. I I

:

e-".

;/{l'dbk ,. .

.&*-&4$k~+

=

.-~ -.:. i"" ~ - ~:A- ~-..-- ~ L~ a-& ~ ~ ~~L.=. ~. .~ ~ =~ ~Vcoii

-.>-

I

-

.., Simulation m u l l s far EPI waveform with 1200V drop on the ~

.Y

. I

0-

..I

.I..

.I..

0s..

inductance with modified control

In order to keep the balance of the energy, a modification to the algorithm is proposed. The two high voltage bridges need to be pulse width modulated for two voltage ranges: from 400V to SOOV and from 1200V to 1600V. The modified control logic is listed in Table 11, and the relationship between the control voltage and voltage supplied by each bridge is shown in Fig.10. The idea is to keep the duty cycle of the 400V bridge unchanged, but with a different output polarity on the critical voltage range boundaries. In this mode the voltage is realized as: 800.(1- 0)- 4 0 0 . 0 = 4 0 0 . 0 , for the 400-8OOV range, and 800+800.(1- 0)-400.0 = 800+ 400. 0 .for the 1200-16OOV range.

Fig. 12 Control block diagram

bridges are positive. The new modulation proposed for this option covers the ,~ common .. ~ ~ .waveforms ~ ~ ~ . used . ~ for ~ conventional ~ . . MRI gradients. However, it does not warranty correct operation for any possible waveform, for example a shmvtooth shaped current could require increasing the coil current with high voltage and reducing it with low voltage. This will result in discharging the coil charged from the high voltage supplies into the low voltage supply, creating a possible over voltage situation. This issue would have to be addressed by regulating the power supplies outputs by the selection of the modulation, or a large resistor to dissipate the extra energy. The previously mentioned issues, together with low reduction in switching losses because of the need of modulating the duty cycle of the high voltage bridges, makes this option less desirable than option 1. 111.

EXPERIMENTAL CIRCUIT

The amplifier was implemented using Option I control. The voltages used in the experimental prototype were 200V, and 700V for the low and high voltages respectively, because of power supply availability in the amplifier modified for this feasibility prototype. Some modifications were implemented in the control to achieve better performance. Figure 12 shows the block diagram of the control. The low voltage is used for the feedback regulation and the resistive voltage drop of the coil. In case a higher voltage than 1400V is needed for the coil current slew rate the additional voltage is obtained from the low voltage bridge. Also, the bridges input voltage is sensed and the feedforward is calculated with the actual values, increasing

264

2004 35rh A n n u l IEEE Power Elecrronics Specialists Conference

Aachen. Germany, 2004

5flflV/div., time 4flflusldiu.)

Fig. I S Error with respect to the desired value

significantly the rejection of input voltage variation due to power supply regulation transients. The power stage of the prototype was implemented with Powerex CM600HU24H IGBTs for all the bridges, which were available in the amplifier modified for this prototype. The final implementation will use 600V devices for the low voltage bridge for lower losses. The prototype was built with the purpose to evaluate the control technique selected and to assess the performance, not to verify the maximum current achievable with a fully optimized design. The control was implemented using a TMS3206716 @720MHz DSP, and a Xilinx FPCA. Figure 13 shows a block diagram of the hardware implementation. The results obtained with the prototype are shown in the next figures. Figure 14 shows the voltage and current when driving the coil with a trapezoidal current waveform of +50A amplitude and 0.3 N u s slew rate, with Ims flat top. Figure 15 shows the error during transients with respect to the desired values, the peaks are on the range of one amp corresponding to the corners of the waveform and much smaller values for the ramp and the flat top. Figure 16 shows details of the current and voltage arround the comer transitions for a 20011s flat top case. Figure 17 shows the case for a spiral waveform with a zoom detail. The error for this case is shown in Fig. 18. The results obtained verify the performance expectation

2flA/d&., Voltage SflflVIdiv., time 200us/div.;Bottom: lcoil IOA/div., Voltage 25flV/div., time 2flusIdiv.) Fig. 16 Detail of voltage and current for lhe coil for a fSflAtrapezoidal waveform with a2flflur flat top

for the prototype, and no overvoltage at any power supply output was detected. It is especially good to point out the performance during a slope change. The digital control made possible to use a calibrated feedforward, which allows replicating features that have much higher frequency contents than the control bandwidth. The operation with only feedback control would not have resulted in such a performance.

1v.

CONCLUSION

The proposed power stage and its control provide fast dynamic performance and high power using the high bandwidth of a low voltage high switching frequency bridge stacked with two high voltage low switching frequency bridges. Two possible control methods have been analyzed. The control with lower switching losses has Iimitations on the type of waveforms it can replicate. Possible solutions have been proposed to prevent the power supply overvoltage for this modulation technique. The control method with no restrictions on the waveforms is the preferred option in spite

265

Aachen, Germany, 2004

2004 35th Annual IEEE Power Electronics Specialists Conference

The selected power stage and control method for gradient amplifier provides the high power and high current waveform accuracy with the required performance for MRI systems. The arbitrary waveform capability not only provides high performance for current standard scanning techniques, but will also enable new techniques with more complex current waveform shapes. REFERENCES

[I]

of higher switching losses, because of its arbitrary waveform capability and expected high bandwidth. The selected control has been implemented in a feasibility

prototype. The experimental results obtained confirm the expected

performance. The combination of feedback and feedforward using a digital control make possible to replicate waveform features that have frequency contents higher than the control loop bandwidth.

Robert L. Steigenvald, William F. Wirth, “High-Power, High-

Performance Switching Amplifier Driving Magnetic Resonance Imaging Gradient Coils,” IEEE PESC 2000 Con/ Proceedings, pp.64347. [2] S . Watanabe, P. Boyagoda, H. lwamoto, M. Nakaoka, H. Takano.”Pawer Conversion PWM Amplifier with Two Paralleled Four Quadrant Chopper for MRI Gradient Coil Magnetic Field Current Tracking Implementation,” IEEE PESC 1999 Con! Proceedings. pp.909-913. 131 S . Wamabe, P. Boyagoda, H. Iwamoto, M. Nakaoka, H. Takmo. “Analysis on a PWM Power Conversion Amplifier with IGBT Macro Model to Generate Gradient Magnetic.Fields in MRI Systems,” IEEE PEDS1999Conf Record, pp.127-132. [4] Stanley, G.R.; Bradshaw, K.M.; “Precision E€-to-AC power conversion by optimization of the output current waveform-the half bridge revisited,” IEEE Trans. on Power Elecrronics. Vol. I4 Iss. 2 , March 1999, pp. 372 -380 [SI H. Fukuda, M. Nakaoka, “High-Frequency ZVS PWM Power Amplifier for Magnetic-Field Current Tracking Control Scheme and its Design Considerations,” IEEE PEDS Con/ Record 1995, pp. 238245. [ 6 ] R.S. Bunven.“Kilowalts on Order (Power Amplifiers),” IEEE Specmum February 1993. Vol.: 3 0 1 s ~2, . pp. 32-37. [7] O.M. Mueller, J.N. Park, P.B. Roemer, S.P. S o u q “A HighEfficiency 4-Switch GTO Speed-up Inverter for the Generation of Fast-Changing MRI Gradient Fields,” IEEE APEC Confireme Proceedings 1993. pp. 806-812. [SI O.M. Mueller, P.B. Roemer, J.N. Park, S.P. Souza, R.L. SI. Peters. “A GTO Speed-Up Inverter for Fast-Scan Magnetic Resonance Imaging,” IEEE APEC Conference Proceedings 1992. pp. 479.486. 191 R. Steigenvald, W.F.Wirth. “High-Efficiency Switching Power Amplifier,” US Parenr. (iS6,166,602. Dec. 26, 2W0. [IO] W.F. Wirth, T.G. McFarland, R. M. Vavrek, P. Rocmez O.M. Mueller, J.N. Park, “Split Gradient Amplifier for an MRI System” US Parent. USS.270.657,Dec. 14, 1993. [ I l l P. Schweighofer, ‘‘Method far the Generation of Control Signals for a Power Amplifier and a Power Amplifier,’’ US Patent. US 6,118,337, sep.12. 2000. 1121 J.A. Sabate, M. Schutlen; Qiming Li; W.F. Wirth. “Resonant power supply for magnetic resonance imaging gradient drivers.” IEEE PESC ’03 Conference Proceedings 2003, pp. 1815 -1820.

266