Token ring based spike scan arbitration circuit
By using a token ring-based spike scanning arbitration circuit, the problems of slow arbitration speed and poor fairness in dynamic vision sensors are solved, improving arbitration speed and fairness, and enhancing data throughput and imaging quality.
Patent Information
- Application Number
- CN202310184860.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-03-01
- Publication Date
- 2026-02-06
- Estimated Expiration
- 2043-03-01
AI Technical Summary
The arbitration circuits of existing dynamic vision sensors suffer from slow arbitration speed and poor arbitration fairness, resulting in reduced data throughput and decreased image quality.
A token ring-based spike scanning arbitration circuit is adopted, which includes a spike scanning circuit and a token ring circuit. The spike scanning circuit scans the most significant bits of the request signal sequentially, and the token ring circuit realizes the priority of the request signal to be rotated in time and space, thereby improving the arbitration speed and fairness.
This enables faster arbitration speed and fairer arbitration for dynamic vision sensors, improving data throughput and image quality.
Smart Images
Figure CN116233633B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of electronic circuits, and relates to an arbitration circuit, in particular to a token ring-based spike scanning arbitration circuit in the field of microelectronics, which can be used for a dynamic vision sensor. BACKGROUND
[0002] The dynamic vision sensor comprises a pixel circuit, a line or circuit, an arbitration circuit, an encoding circuit and a handshake circuit. The arbitration circuit is responsible for bus communication of the dynamic vision sensor and solving signal conflict problems, and ensures accurate and rapid processing of logic signals of the pixel circuit. Arbitration speed and arbitration fairness of the arbitration circuit are the most important two performances of the arbitration circuit, and the arbitration speed determines the data throughput of the dynamic vision sensor chip, and the data throughput and the arbitration fairness directly affect the imaging quality.
[0003] In 2022, Zhang Zhiyuan disclosed an average arbitration circuit applied to a dynamic vision sensor and composed of cascaded two-input average arbitration circuits in his "Bionic Dynamic Vision Sensor Integrated Circuit Design" (Zhang Zhiyuan. Xi'an University of Electronic Science and Technology, 2020.), the structure of which is shown in Figure 1 , the resolution of the dynamic vision sensor is 2 N ×2 N , and the average arbitration circuit of a row or a column is composed of 2 N -1 cascaded two-input average arbitration circuits, wherein N≥1; the structure of the two-input average arbitration circuit is shown in Figure 2 , which is composed of a preselection unit, a priority selection unit and a communication unit. When the input is 2 N , and the arbitrator requests at the same time when 2 N request signals occur, the arbitrator can realize the rotation of the priority of each request signal.
[0004] The average arbitration circuit needs to arrange 2 N -1 two-input average arbitration circuits in a tree with N levels, and arbitrate between 2 N request signals; the arbitration circuit of the root level selects the winning sub-tree, and the signal propagates downward along the tree. This process continues until the input obtains confirmation of access to the shared output bus. When multiple requests arrive in a short time, one of the request signals will win the arbitration, and all other requests will be queued until the winning input releases the output bus.
[0005] The average arbitration circuit can rotate the priority of all input request signals to realize fair arbitration, but it has defects that M(M≥2) request signals must be propagated and arbitrated in each layer of the average arbitration circuit, each request signal passes through log2M stages, and the total propagation time is Mlog2M, when the dynamic visual sensor works in a large number of event environment, the delay generated by the arbitration circuit rises exponentially, which reduces the arbitration speed of the circuit, thereby reducing the data throughput of the dynamic visual sensor, and further reducing the imaging quality of the dynamic visual sensor. Moreover, for the average arbitration circuit, although the priority of all input request signals can be rotated, this fairness presents sparse characteristics in space and time, which does not conform to the motion law of objects in the actual detection scene, and affects the imaging quality of the dynamic visual sensor. SUMMARY
[0006] The present application aims at the deficiencies of the prior art, and provides a spike scanning arbitration circuit based on a token ring, to solve the technical problems of slow arbitration speed and poor arbitration fairness in the prior art.
[0007] To achieve the above object, the technical scheme adopted by the present application is as follows:
[0008] A spike scanning arbitration circuit based on a token ring, comprising a spike scanning circuit and a token ring circuit; the spike scanning circuit comprises a first complement unit, a preprocessing unit composed of a subtraction unit and an inverter INV connected in sequence, and an AND gate AGND and an OR gate OR; one input end of the AND gate AGND is connected with the output end of the inverter INV, and the other input end is connected with the output end of the first complement unit, and the output end of the AND gate AGND is connected with a high-bit selection unit and a low-bit selection unit arranged in parallel; the two input ends of the OR gate OR are connected with the high-bit selection unit and the low-bit selection unit respectively; the token ring circuit comprises an edge detection unit, a token unit and a second complement unit connected in sequence; the input end of the edge detection unit is connected with the output end of the OR gate OR, and the output end of the second complement unit is connected with the subtraction unit;
[0009] The first complement unit is used for complementing the request signal req[0:N-1] input from the outside with N high bits to generate a preprocessing request signal req2[0:2N-1], and N≥2;
[0010] The pre-processing unit, wherein the input positive terminal of the subtracting unit connected with the first padding unit is used to subtract the pre-processing request signal req2[0:2N-1] from the padding token signal token1[0:2N-1] output by the second padding unit bit by bit, the inverter INV is used to invert the signal minus[0:2N-1] after bit by bit subtraction to generate the priority selection signal minus_inv[0:2N-1];
[0011] The AND gate AGND is used to perform AND operation on the pre-processing request signal req2[0:2N-1] and the minus_inv[0:2N-1] to generate the pre-processing response signal grant2[0:2N-1];
[0012] The high-bit selection unit and the low-bit selection unit are respectively used to select the N-bit high pre-processing response signal grant2[N:2N-1] containing one high-bit target spike information of req[0:N-1] in the pre-processing response signal grant2[0:2N-1] and the N-bit low pre-processing response signal grant2[0:N-1] containing one low-bit target spike information of req[0:N-1];
[0013] The OR gate OR is used to perform OR operation on grant2[0:N-1] and grant2[N:2N-1] bit by bit to generate the response signal grant[0:N-1] of the spike with the high effective bit of req[0:N-1];
[0014] The edge detection unit is used to detect the rising edge of the response signal grant[0:N-1];
[0015] The token unit is used to generate the N-bit token signal token[0:N-1], and when the rising edge of grant[0:N-1] is detected, the high effective bit in the token signal token[0:N-1] is moved to the high bit by one bit;
[0016] The second padding unit is used to pad 0 to the high N bit of the token signal token[0:N-1] to generate the padding token signal token1[0:2N-1].
[0017] The initial value of the pre-processing token signal token1[0:2N-1] output by the second padding unit is a 2N-bit signal 0000…0001.
[0018] The initial value of the token signal token[0:N-1] output by the token unit is an N-bit signal 0000…0001.
[0019] The high-bit target peak information and the low-bit target peak information are req[N-1] and req[0:N-2] in the request signal req[0:N-1] respectively.
[0020] Compared with the prior art, the present application has the following advantages:
[0021] Firstly, the present application contains a peak scanning circuit, which sequentially scans the high effective bits of the request signal req[0:N-1], selects the first high effective bit from the scanning starting point, and propagates arbitration in one layer. When M input request signals come at the same time, M response signals are outputted after M times of request propagation, each request only goes through one stage, the total propagation time is M, the average processing time of a single request signal is reduced by log2M times, the speed of the arbitration circuit is improved, the large data throughput of the dynamic vision sensor can be realized, and thus the imaging quality is improved.
[0022] Secondly, the present application adopts a token ring circuit, the request signal corresponding to the high effective bit of the token signal is the starting point of the peak scanning, the initial value of the token signal is 0000…0001, and when the arbitration circuit outputs a high effective response signal, the rising edge of the response signal is detected by the edge detection unit, the high effective bit of the token signal moves one bit to the high bit. When the high effective bit of the token signal is at the highest bit, the next high effective bit returns to the lowest bit, the arbitration priority of all signals is rotated in time and space, the fairness is continuous in time and space, which conforms to the motion law of the target object in the actual application of the dynamic vision sensor, the arbitration of the arbitration circuit is more fair, and the imaging quality of the dynamic vision sensor is improved. BRIEF DESCRIPTION OF DRAWINGS
[0023] Figure 1 Fig. 2 is a structural schematic diagram of a two-input average arbitration circuit cascaded by the average arbitration circuit of the prior art;
[0024] Figure 2 Fig. 2 is a structural schematic diagram of a two-input average arbitration circuit cascaded by the average arbitration circuit of the prior art;
[0025] Figure 3 Fig. 2 is a structural schematic diagram of a two-input average arbitration circuit cascaded by the average arbitration circuit of the prior art; DETAILED DESCRIPTION
[0026] The present application will be further described in detail below in combination with the drawings and specific embodiments.
[0027] Reference Figure 3 The present application includes a peak scanning circuit and a token ring circuit, N request signals form a request signal req[0:N-1], and N≥2.
[0028] The peak scanning circuit comprises a first padding unit, a preprocessing unit composed of a subtraction unit and an inverter INV connected in sequence, and an AND gate AGND and an OR gate OR for scanning the high effective bits of the external input request signal req[0:N-1] in sequence, i.e. scanning the high effective bits of the request signal req[0:N-1] in sequence, and selecting the first high effective bit from the scanning starting point, wherein:
[0029] The first padding unit pads N bits of 0 to the high bits of req[0:N-1] to generate a preprocessed request signal req2[0:2N-1], and the padding of N bits of 0 ensures that the processing of the subtraction unit does not result in the loss of the high effective bit data of the request signal.
[0030] The subtraction unit subtracts the preprocessed request signal req2[0:2N-1] from the padding token signal token1[0:2N-1] output by the second padding unit bit by bit to generate a signal minus[0:2N-1].
[0031] The inverter INV inverts the signal minus[0:2N-1] bit by bit to generate a priority selection signal minus_inv[0:2N-1].
[0032] The AND gate AGND performs an AND operation on the preprocessed request signal req2[0:2N-1] and the minus_inv[0:2N-1] to generate a 2N-bit preprocessed response signal grant2[0:2N-1].
[0033] The high-bit selection unit and the low-bit selection unit are respectively used to select the high N-bit preprocessed response signal grant2[N:2N-1] and the low N-bit preprocessed response signal grant2[0:N-1] of the preprocessed response signal grant2[0:2N-1]; and the OR gate OR performs a bitwise OR operation on grant2[0:N-1] and grant2[N:2N-1] to generate an N-bit response signal grant[0:N-1] to be output to the outside world, and the grant[0:N-1] has at most one high effective bit, and when there is one high effective bit, it indicates that the arbitration circuit responds to the request signal and generates a corresponding response signal.
[0034] The peak scanning circuit generates a response signal corresponding to the request signal in one stage, and compared with the hierarchical arbitration of the binary tree arbitration circuit, the arbitration speed is faster.
[0035] The token ring circuit is used to generate a token signal token[0:N-1], and when the rising edge of grant[0:N-1] is detected, the high valid bit in the token signal token[0:N-1] is moved to the high bit by one bit, and the high valid bit position of token[0:N-1] is the scanning starting point of the spike scanning circuit, wherein:
[0036] The edge detection unit is used to detect the rising edge of the response signal grant[0:N-1].
[0037] The token unit generates an N-bit token signal token[0:N-1], and when the rising edge of grant[0:N-1] is detected, the high valid bit in the token signal token[0:N-1] is moved to the high bit by one bit, a new token signal token[0:N-1] is generated, the initial value of token[0:N-1] is an N-bit signal 0000…0001, only the lowest bit is 1, and the remaining bits are 0. The token signal moves sequentially in space, and the request signal is fairly allocated priority, which meets the space-time characteristics of the target object motion.
[0038] The second bit supplementing unit supplements N bits of 0 to the high bit of the token signal token[0:N-1] to generate a 2N-bit preprocessed token signal token1[0:2N-1], and the high N bits are supplemented with 0 to ensure that the preprocessed request signal req2[0:2N-1] and the preprocessed token signal token1[0:2N-1] are both 2N bits, which can be subtracted.
[0039] The working principle of the application is that the high effective bit in the token signal token[0:N-1] generated by the token unit in the token ring circuit determines the starting point of the spike scanning, the spike scanning circuit scans the high effective bit of the external input request signal req[0:N-1] from the starting point in turn, and determines the nearest high effective bit from the scanning starting point, and outputs the corresponding response signal grant[0:N-1] to the outside world; the edge detection unit determines whether the request-response is completed by detecting the rising edge of each bit of the response signal grant[0:N-1], and after the completion, the high effective bit in the token signal token[0:N-1] generated by the token unit moves 1 bit to the high bit. The high effective bit of the request signal req[0:N-1] is scanned in turn by the spike scanning circuit, and the first high effective bit from the scanning starting point is selected, and the request signal is only propagated in one layer for arbitration, which improves the speed of the arbitration circuit and can realize the large data throughput of the dynamic vision sensor. The request signal corresponding to the high effective bit in the token signal is the starting point of the spike scanning, the initial value of the token signal is 0000…0001, and whenever the arbitration circuit outputs a high effective response signal, the rising edge of the response signal is detected by the edge detection unit, and the high effective bit of the token signal moves one bit to the high bit. When the high effective bit of the token signal is in the highest bit, the next high effective bit returns to the lowest bit, realizing that the arbitration priority of all signals is rotated in turn in time and space. This fairness is continuous in time and space, which conforms to the motion law of the target object in the actual application of the dynamic vision sensor, and realizes the more fair arbitration of the arbitration circuit. The faster arbitration speed and the more fair arbitration improve the imaging quality of the dynamic vision sensor.
Claims
1. A spike scanning arbitration circuit based on a token ring, characterized in that, The system includes a spike scanning circuit and a token ring circuit. The spike scanning circuit includes a first complement unit connected in sequence, a preprocessing unit consisting of a subtraction unit and an inverter INV connected in sequence, and an AND gate AGND and an OR gate OR. One input of the AND gate AGND is connected to the output of the inverter INV, and the other input is connected to the output of the first complement unit. The output of the AND gate AGND is connected to a high-order selection unit and a low-order selection unit arranged in parallel. The two inputs of the OR gate OR are respectively connected to the high-order selection unit and the low-order selection unit. The token ring circuit includes an edge detection unit, a token unit, and a second complement unit connected in sequence. The input of the edge detection unit is connected to the output of the OR gate OR, and the output of the second complement unit is connected to the subtraction unit. The first padding unit is used to pad the high N bits of the externally input request signal req[0:N-1] with 0, and generate a preprocessed request signal req2[0:2N-1], where N≥2; The preprocessing unit, wherein the positive input terminal is connected to the first padding unit, is used to subtract the preprocessing request signal req2[0:2N-1] from the padding token signal token1[0:2N-1] output by the second padding unit bit by bit, and the inverter INV is used to invert the bit-subtracted signal minus[0:2N-1] bit by bit to generate the priority selection signal minus_inv[0:2N-1]. The AND gate AGND is used to perform an AND operation on the preprocessing request signal req2[0:2N-1] and minus_inv[0:2N-1] to generate the preprocessing response signal grant2[0:2N-1]. The high-bit selection unit and the low-bit selection unit are respectively used to select an N-bit high-bit preprocessed response signal grant2[N:2N-1] containing one high-bit target spike information of req[0:N-1] and an N-bit low-bit preprocessed response signal grant2[0:N-1] containing one low-bit target spike information of req[0:N-1]. The OR gate is used to perform a bitwise OR operation on grant2[0:N-1] and grant2[N:2N-1] to generate a spike response signal grant[0:N-1] with the most significant bit being req[0:N-1]. The edge detection unit is used to detect the rising edge of the response signal grant[0:N-1]. The token unit is used to generate an N-bit token signal token[0:N-1]. When the rising edge of grant[0:N-1] is detected, the most significant bit in the token signal token[0:N-1] is shifted one bit to the most significant bit. The second padding unit is used to pad the high N bits of the token signal token[0:N-1] with 0 to generate the padding token signal token1[0:2N-1].
2. The token ring-based spike scanning arbitration circuit according to claim 1, characterized in that, The second padding unit outputs a preprocessed token signal token1[0:2N-1] with an initial value of 2N bits, 0000…0001.
3. The token ring-based spike scanning arbitration circuit according to claim 1, characterized in that, The token unit outputs a token signal token[0:N-1] with an initial value of N bits: 0000…0001.
4. The token ring-based spike scanning arbitration circuit according to claim 1, characterized in that, The high-level target spike information and the low-level target spike information refer to req[N-1] and req[0:N-2] in the request signal req[0:N-1].
Citation Information
Patent Citations
High-speed asynchronous interlinkage communication network of heterogeneous multi-nucleus processor
CN101132336A
QC-LDPC (Quasi-Cyclic Low Density Parity Check) code constructing method
CN101826878A