Modular pulse synthesizer for transcranial magnetic stimulation with fully adjustable pulse shape and sequence

Abstract

The temporal shape of a pulse in transcranial magnetic stimulation (TMS) influences which neuron populations are activated preferentially as well as the strength and even direction of neuromodulation effects. Furthermore, various pulse shapes differ in their efficiency, coil heating, sensory perception, and clicking sound. However, the available TMS pulse shape repertoire is still very limited to a few biphasic, monophasic, and polyphasic pulses with sinusoidal or near-rectangular shapes. Monophasic pulses, though found to be more selective and stronger in neuromodulation, are generated inefficiently and therefore only available in simple low-frequency repetitive protocols. Despite a strong interest to exploit the temporal effects of TMS pulse shapes and pulse sequences, waveform control is relatively inflexible and only possible parametrically within certain limits. Previously proposed approaches for flexible pulse shape control, such as through power electronic inverters, have significant limitations: The semiconductor switches can fail under the immense electrical stress associated with free pulse shaping, and most conventional power inverter topologies are incapable of generating smooth electric fields or existing pulse shapes. Leveraging intensive preliminary work on modular power electronics, we present a modular pulse synthesizer (MPS) technology that can, for the first time, flexibly generate high-power TMS pulses (one-side peak ∼4000 V, ∼8000 A) with user-defined electric field shape as well as rapid sequences of pulses with high output quality. The circuit topology breaks the problem of simultaneous high power and switching speed into smaller, manageable portions, distributed across several identical modules. In consequence, the MPS TMS techology can use semiconductor devices with voltage and current ratings lower than the overall pulse voltage and distribute the overall switching of several hundred kilohertz among multiple transistors. MPS TMS can synthesize practically any pulse shape, including conventional ones, with fine quantization of the induced electric field (⩽17% granularity without modulation and ∼300 kHz bandwidth). Moreover, the technology allows optional symmetric differential coil driving so that the average electric potential of the coil, in contrast to conventional TMS devices, stays constant to prevent capacitive artifacts in sensitive recording amplifiers, such as electroencephalography. MPS TMS can enable the optimization of stimulation paradigms for more sophisticated probing of brain function as well as stronger and more selective neuromodulation, further expanding the parameter space available to users.

Keywords: activation selectivity; arbitrary waveform generation; electric field quantization; flexible pulse shape synthesis; pulse sequence control; temporal control flexibility; transcranial magnetic stimulation.

Figure 1.

Circuit topologies (left column) and corresponding typical pulse shapes (right column) of various TMS technologies: A. Monophasic stimulator (typically with thyristors), B. bi- and polyphasic stimulator (existing with thyristors and IGBTs), C. example of a bridge-based TMS device using IGBTs, representing also all other bridge approaches, and D. modular pulse synthesizer. This work implements six four-bridge modules (double H bridge) as shown on the right, which the control can intrinsically operate also as two-bridge modules (H bridge, shown as alternative on the left and compared in Table I). The topology can use chargers and dischargers in several positions. Although every module could have its own charger, a single one in combination with the circuit’s charge distribution capabilities is sufficient and can reduce commonmode currents through the parasitic and filter capacitances of the chargers, which are relevant for electromagnetic interference.

Figure 2.

A. Circuit topology of the power train with here four four-bridge modules. B. Example for a state of the system with two modules including their internal capacitor connected in series and two in parallel to generate two output steps. Alternative two-bridge modules operate similarly, but the parallel mode would have to be replaced by bypass. See Figure 3 for definition of all states.

Figure 3.

Switch states of an interconnection A between two modules or B the outmost modules connecting to the coil. The states influence how the capacitors of the modules are electrically connected relative to their neighbors. Two-bridge modules can generate only series and bypass states, four-bridge modules additionally parallel circuit configurations between modules. Series states increase the voltage in positive or negative direction. Parallel modes distribute the current load and balance charge across the capacitors. With fast transistors, the transition from one state to another can occur on the order of 100 ns and several hundred times per millisecond.

Figure 4.

Principle of phase-shifted carrier modulation (PSC) with here four modules M_k for simplicity and accordingly four carriers C_k. to reproduce the black reference curve m(t). A. Desired reference shape and carrier for comparison, B. module switching states s_k for every module k resulting from the comparison, and C. quantized output following from this modulation.

Figure 5.

Simulation results (A) without compensation and (B) with a reference compensating resistive voltage drop and module capacitor voltage variations. The model does not use the parallel mode but only series and bypass to demonstrate how the capacitors drift apart. Parallelization clears such voltage differences.

Figure 6.

Simulation results using a reference that compensates the resistive voltage drop, module capacitor voltage variations, and the final module voltage spread.

Figure 7.

Circuit board layout (A. actual photo and B. design) with switching cells in the front and capacitance in the back. The transistors are on the bottom side, the driver and first-stage capacitance on the top.

Figure 8.

Two different assignments of four switching cells (discrete pair of high-side and low-side transistor) to two output bridges on one side of a module demonstrating the balance of the current among discrete parallel switching cells. Each switching cell contains discrete transistors (on the back, not visible), gate driver (black ICs between the golden islands), and first-level ceramic capacitance (yellow blocks in V configuration) to absorb the immediate energy demand and/or supply when the parasitic magnetic fields of the commutation current paths is discharged or charged.

Figure 9.

Double-pulse setup.

Figure 10.

Example for one of four commutation conditions for A the original FF11 transistor bridge in the optimized layout and B with an optimized package for lower stray inductance. In the measurements, one transistor module is singled out.

Figure 11.

Temperature development of the SiC transistors under stress testing with sinusoidal current (700 A peak) per discrete transistor (enabling 8,400 A for a module with 12 transistors in our system) and transistor switching rates up to 100 kHz. Dashed lines were performed without any additional cooling in a closed 19” cabinet, solid lines included active air convection, demonstrating a wide margin. The temperature development and the steady-state temperatures show a widely linear relationship with the load as expected. The naming convention is pulse rate (Hz)/pulse duration /switching rate during the pulses.

Figure 12.

Recordings of biphasic pulses with cosinusoidal voltage as well as electric field shape and sinusoidal current shape. The pulse shape including its duration is purely software-generated (A 200 μs, B 400 μs, C 600 μs).

Figure 13.

Recorded pulse train of typical conventional pulses (biphasic and two half-period pulses) with as little as 50 μs between two entirely different pulse shapes. The pulses have different energy content. Still, the device does not need any reconfiguration time or adjustment of charge to change the pulse duration, the pulse amplitude, or change the entire pulse shape. The panels from A to D show the same pulse sequence with increasing pulse amplitude. The recordings demonstrate both ways how the pulse synthesizer can adjust the pulse amplitude: Within a sequence, the circuit uses fewer steps for the lower-amplitude sections and pulses such as the long half-period pulse in the center. From pulse train to pulse train, the circuit increases the amplitude through increasing the module voltage step size.

Figure 14.

Sequence of monophasic pulses generated with various pulse durations and in two amplitudes. The approximately exponentially falling phase of the current in the second part of each pulse is generated through resistive damping in conventional monophasic TMS devices associated with the loss of the entire pulse energy. The pulse synthesizer, however, simulates only a resistor to the coil and extracts the energy back to the internal capacitors through proper control so that there is sufficient energy to generate such rapid sequences, which no conventional monophasic device can. Whereas we demonstrated the adjustment of the module voltages to tune the pulse strength in Figure 13, here the lower amplitude of the sequence in Panel A uses less module levels than Panel B to clarify the coarser quantization of this approach.

Figure 15.

Burst of various monophasic pulses with different amplitude and time direction. To demonstrate the flexibility and ability to generate practically any pulse shape, the burst does not only change the amplitude but inverses the time direction for the third and fourth pulse so that the voltage is mirrored and the current additionally inverted.

Figure 16.

Polyphasic pulses with sinusoidal current and Gaussian envelope of increasing width from A to D. For very short envelope width, the pulse degenerates into a near-sinusoidal biphasic pulse, which in contrast to conventional biphasic TMS pulses contains smoother edges in the voltage.

Figure 17.

Polyphasic pulses with cosinusoidal current and Gaussian envelope of increasing width from A to D. For very short envelope width, the pulse degenerates into a near-sinusoidal biphasic pulse, which in contrast to conventional biphasic TMS pulses contains smoother edges in the voltage.

Figure 18.

Pulse shapes optimized for minimum coil heating with different pulse duration (A and B) and various pulse amplitudes for the longer duration (B – E). Before the actual pulse, the device brings the current baseline slowly to a negative offset, from which the current can ramp up with a longer rising edge without reaching large currents. This pre-phase is generated through subtle control of the voltage in the approximately 120 μs introducing the pulse.

Figure 19.

Unfiltered voltage and current spectra of (A) a biphasic pulse, which may mostly be dominated by an oscillation with one dominant frequency but inherently displays a wide spectrum to its brevity (frequency–time uncertainty principle), and (B) a polyphasic pulse with more periods and Gaussian envelope for a sharper spectrum of the intended pulse. With a careful look, one can see the small residual distortion due to both measurement noise and technical quantization emerging at the right side of the voltage spectrum, in the high-frequency range. The inset in each panel shows the corresponding time course of the voltage or current.