Suraj Gowda

Powering and communicating with mm-size implants

This paper deals with system level design considerations for mm-size implantable electronic devices with wireless connectivity. In particular, it focuses on neural sensors as one application requiring such miniature interfaces. Common to all these implants is the need for power supply and a wireless interface. Wireless power transfer via electromagnetic fields is identified as a promising option for powering such devices. Design methodologies, system level trade-offs, as well as limitations of power supply systems based on electromagnetic coupling are discussed in detail. Further, various wireless data communication architectures are evaluated for their feasibility in the application. Reflective impulse radios are proposed as an alternative scheme for enabling highly scalable data transmission at <1pJ/bit. Finally, design considerations for the corresponding reader system are addressed.

Continuous closed-loop decoder adaptation with a recursive maximum likelihood algorithm allows for rapid performance acquisition in brain-machine interfaces

Closed-loop decoder adaptation (CLDA) is an emerging paradigm for both improving and maintaining online performance in brain-machine interfaces (BMIs). The time required for initial decoder training and any subsequent decoder recalibrations could be potentially reduced by performing continuous adaptation, in which decoder parameters are updated at every time step during these procedures, rather than waiting to update the decoder at periodic intervals in a more batch-based process. Here, we present recursive maximum likelihood (RML), a CLDA algorithm that performs continuous adaptation of a Kalman filter decoders parameters. We demonstrate that RML possesses a variety of useful properties and practical algorithmic advantages. First, we show how RML leverages the accuracy of updates based on a batch of data while still adapting parameters on every time step. Second, we illustrate how the RML algorithm is parameterized by a single, intuitive half-life parameter that can be used to adjust the rate of adaptation in real time. Third, we show how even when the number of neural features is very large, RMLs memory-efficient recursive update rules can be reformulated to also be computationally fast so that continuous adaptation is still feasible. To test the algorithm in closed-loop experiments, we trained three macaque monkeys to perform a center-out reaching task by using either spiking activity or local field potentials to control a 2D computer cursor. RML achieved higher levels of performance more rapidly in comparison to a previous CLDA algorithm that adapts parameters on a more intermediate timescale. Overall, our results indicate that RML is an effective CLDA algorithm for achieving rapid performance acquisition using continuous adaptation.

Adaptive Kalman filtering for closed-loop Brain-Machine Interface systems

Brain-Machine Interface (BMI) decoding algorithms are often trained offline, but this paradigm ignores both the non-stationarity of neural signals and the feedback that exists in online, closed-loop control. To address these problems, we have developed an Adaptive Kalman Filter (AKF), a Kalman filter variant that adaptively updates its model parameters during training. For a Kalman filter decoder, batch retraining methods require completely re-estimating the parameter matrices from sufficient data to perform regression accurately, even if only small changes are necessary. Conversely, the AKF is designed to update the decoder parameters continuously and more intelligently. We simulated a population of 41 neurons learning to control a 2D computer cursor. The AKF yielded significantly faster skill acquisition and better robustness to perturbation and neuron loss than a standard Kalman filter with periodic batch retraining.

Rapid control and feedback rates enhance neuroprosthetic control

Brain-machine interfaces (BMI) create novel sensorimotor pathways for action. Much as the sensorimotor apparatus shapes natural motor control, the BMI pathway characteristics may also influence neuroprosthetic control. Here, we explore the influence of control and feedback rates, where control rate indicates how often motor commands are sent from the brain to the prosthetic, and feedback rate indicates how often visual feedback of the prosthetic is provided to the subject. We developed a new BMI that allows arbitrarily fast control and feedback rates, and used it to dissociate the effects of each rate in two monkeys. Increasing the control rate significantly improved control even when feedback rate was unchanged. Increasing the feedback rate further facilitated control. We also show that our high-rate BMI significantly outperformed state-of-the-art methods due to higher control and feedback rates, combined with a different point process mathematical encoding model. Our BMI paradigm can dissect the contribution of different elements in the sensorimotor pathway, providing a unique tool for studying neuroprosthetic control mechanisms.

Designing dynamical properties of brain-machine interfaces to optimize task-specific performance.

Brain-machine interfaces (BMIs) are dynamical systems whose properties ultimately influence performance. For instance, a 2-D BMI in which cursor position is controlled using a Kalman filter will, by default, create an attractor point that “pulls” the cursor to particular points in the workspace. If created unintentionally, such effects may not be beneficial for BMI performance. However, there have been few empirical studies exploring how various dynamical effects of closed-loop BMIs ultimately influence performance. In this work, we utilize experimental data from two macaque monkeys operating a closed-loop BMI to reach to 2-D targets and show that certain dynamical properties correlate with performance loss. We also show that other dynamical properties represent tradeoffs between naturally competing objectives, such as speed versus accuracy. These findings highlight the importance of fine-tuning the dynamical properties of closed-loop BMIs to optimize task-specific performance.

High-performance brain-machine interface enabled by an adaptive optimal feedback-controlled point process decoder.

Brain-machine interface (BMI) performance has been improved using Kalman filters (KF) combined with closed-loop decoder adaptation (CLDA). CLDA fits the decoder parameters during closed-loop BMI operation based on the neural activity and inferred user velocity intention. These advances have resulted in the recent ReFIT-KF and SmoothBatch-KF decoders. Here we demonstrate high-performance and robust BMI control using a novel closed-loop BMI architecture termed adaptive optimal feedback-controlled (OFC) point process filter (PPF). Adaptive OFC-PPF allows subjects to issue neural commands and receive feedback with every spike event and hence at a faster rate than the KF. Moreover, it adapts the decoder parameters with every spike event in contrast to current CLDA techniques that do so on the time-scale of minutes. Finally, unlike current methods that rotate the decoded velocity vector, adaptive OFC-PPF constructs an infinite-horizon OFC model of the brain to infer velocity intention during adaptation. Preliminary data collected in a monkey suggests that adaptive OFC-PPF improves BMI control. OFC-PPF outperformed SmoothBatch-KF in a self-paced center-out movement task with 8 targets. This improvement was due to both the PPFs increased rate of control and feedback compared with the KF, and to the OFC model suggesting that the OFC better approximates the users strategy. Also, the spike-by-spike adaptation resulted in faster performance convergence compared to current techniques. Thus adaptive OFC-PPF enabled proficient BMI control in this monkey.

Parameter estimation for maximizing controllability of linear brain-machine interfaces

Brain-machine interfaces (BMIs) must be carefully designed for closed-loop control to ensure the best possible performance. The Kalman filter (KF) is a recursive linear BMI algorithm which has been shown to smooth cursor kinematics and improve control over non-recursive linear methods. However, recursive estimators are not without their drawbacks. Here we show that recursive decoders can decrease BMI controllability by coupling kinematic variables that the subject might expect to be unrelated. For instance, a 2D neural cursor where velocity is controlled using a KF can increase the difficulty of straight reaches by linking horizontal and vertical velocity estimates. These effects resemble force fields in arm control. Analysis of experimental data from one non-human primate controlling a position/velocity KF cursor in closed-loop shows that the presence of these force-field effects correlated with decreased performance. We designed a modified KF parameter estimation algorithm to eliminate these effects. Cursor controllability improved significantly when our modifications were used in a closed-loop BMI simulator. Thus, designing highly controllable BMIs requires parameter estimation techniques that carefully craft relationships between decoded variables.

Advanced multi-beam spectrometer for the Green Bank Telescope

A new spectrometer for the Green Bank Telescope (GBT) is being built jointly by the NRAO and the CASPER, University of California, Berkeley. The spectrometer uses 8 bit ADCs and will be capable of processing up to 1.25 GHz bandwidth from 8 dual polarized beams. This mode will be used to process data from focal plane arrays. The spectrometer supports observing mode with 8 tunable digital sub-bands within the 1.25 GHz bandwidth. The spectrometer can also be configured to process a bandwidth of up to 10 GHz with 64 tunable sub-bands from a dual polarized beam. The vastly enhanced backend capabilities will support several new science projects with the GBT.

Control of Redundant Kinematic Degrees of Freedom in a Closed-Loop Brain-Machine Interface

Brain–machine interface (BMI) systems use signals acquired from the brain to directly control the movement of an actuator, such as a computer cursor or a robotic arm, with the goal of restoring motor function lost due to injury or disease of the nervous system. In BMIs with kinematically redundant actuators, the combination of the task goals and the system under neural control can allow for many equally optimal task solutions. The extent to which kinematically redundant degrees of freedom (DOFs) in a BMI system may be under direct neural control is unknown. To address this question, a Kalman filter was used to decode single- and multi-unit cortical neural activity of two macaque monkeys into the joint velocities of a virtual four-link kinematic chain. Subjects completed movements of the chain’s endpoint to instructed target locations within a two-dimensional plane. This system was kinematically redundant for an endpoint movement task, as four DOFs were used to manipulate the 2-D endpoint position. Both subjects successfully performed the task and improved with practice by producing faster endpoint velocity control signals. Kinematic redundancy allowed null movements whereby the individual links of the chain could move in a way that cancels out and does not result in endpoint movement. As the subjects became more proficient at controlling the chain, the amount of null movement also increased. Task performance suffered when the links of the kinematic chain were hidden and only the endpoint was visible. Furthermore, all four DOFs of the joint-velocity control space exhibited task-relevant modulation. The relative usage of each DOF depended on the configuration of the chain, and trials in which the less-prominent DOFs were utilized also had better task performance. Overall, these results indicate that the subjects incorporated the redundant components of the control space into their control strategy. Future BMI systems with kinematic redundancy, such as exoskeletal systems or anthropomorphic robotic arms, may benefit from allowing neural control over redundant configuration dimensions as well as the end-effector.

Accelerating Submovement Decomposition With Search-Space Reduction Heuristics

Objective: Movements made by healthy individuals can be characterized as superpositions of smooth bell-shaped velocity curves. Decomposing complex movements into these simpler “submovement” building blocks is useful for studying the neural control of movement as well as measuring motor impairment due to neurological injury. Approach: One prevalent strategy to submovement decomposition is to formulate it as an optimization problem. This optimization problem is nonconvex and finding an exact solution is computationally burdensome. We build on previous literature that generated approximate solutions to the submovement optimization problem. Results: First, we demonstrate broad conditions on the submovement building block functions that enable the optimization variables to be partitioned into disjoint subsets, allowing for a faster alternating minimization solution. Specifically, the amplitude parameters of a submovement can typically be fit independently of its shape parameters. Second, we develop a method to concentrate the search in regions of high error to make more efficient use of optimization routine iterations. Conclusion: Both innovations result in substantial reductions in computation time across multiple nonhuman primate subjects and diverse task conditions. Significance: These innovations may accelerate analysis of submovements for basic neuroscience and enable real-time applications of submovement decomposition.