66What is ONIX?
77==========================================
88ONIX is a data acquisition system for neuroscience, composed of various pieces
9- of hardware. ONIX differs from other acquisition systems in three major points :
9+ of hardware. ONIX differs from other acquisition systems in three ways :
1010
11- 1. Standards
11+ 1. Standards & Interoperability
1212--------------------------------
1313All acquisition systems follow specific sets of rules that outline how data is
1414structured and transmitted between parts of the system. For instance, Intan
@@ -32,21 +32,21 @@ money purchasing separate acquisition systems for each extra tool they wish to
3232add to their experiment. Additionally, for those who want to develop new tools
3333to study the brain, ONIX provides a powerful hardware backend and software
3434infrastructure that can be reused to control almost any type of recoding
35- instrument.
35+ instrument.
3636
37- 2. Tethers
38- --------------------------------
37+ 2. Thin Tethers & Zero Torque Commutation
38+ -------------------------------------------
3939There is a growing appreciation of experiments that examine the natural
4040behaviours of animals. This often means using larger and more intricate
4141(perhaps 3D) experimental setups. It also means that the animal should be
4242impaired as little as possible by the recording setup. To achieve this it is of
4343course important to reducing the weight of the headstage, but the weight of the
4444tether that connects the headstage to the acquisition system is often
45- overlooked. As the animal explores the arena, the centre of mass of the tether
45+ overlooked. As the animal explores the arena, the center of mass of the tether
4646is rarely directly above the animal. Instead, it is off to one side,
47- introducing a rotational force on the animal. The animal must compensate for
48- this torque in order to keep its head up straight. Because the ONI
49- specification allows communication over a single wire, ONIX uses a single
47+ introducing a rotational force (torque) on the animal. The animal must
48+ compensate for this torque in order to keep its head up straight. Because the
49+ ONI specification allows communication over a single wire, ONIX uses a single
5050coaxial cable, making ONIX tethers lighter and thinner compared to classic
5151acquisition systems. ONIX tethers are 0.1 to 0.4 mm in diameter and are
5252extremely flexible.
@@ -56,28 +56,37 @@ becoming twisted. Commutators are hardware devices that allow the tether to
5656rotate while maintaining an electrical connection to the rest of the path to
5757the acquisition system. As the animal moves through the arena, the ONIX
5858commutator receives 3D tracking information from the headstage and drives a
59- motor to actively rotate the tether, preventing twisting.
59+ motor to actively rotate the tether, preventing twisting. Importantly, this
60+ process is driven by real-time measurement of headstage rotation, *not by
61+ torque transmitted by the tether *. This allows nearly zero-torque commutation
62+ and the use of tethers that are so thin, they would not not function in systems
63+ that require torque measurements to drive active commutation.
6064
61- 3. Latencies
65+ 3. Low Latencies
6266--------------------------------
6367In closed-loop experiments, data is not only acquired, but also processed and
6468acted upon. For instance, one can provide optogenetic stimulation to a brain
6569area every time a certain type of event is detected by an extracellular probe.
6670The closed-loop latency of the acquisition system describes how much time
67- passes between the initial event and the response of the system. In classic
68- acquisition systems, this time is primarily spent on transmitting and
69- processing acquired data, which becomes more and more challenging as the number
70- of channels on a probe increases. A short latency allows the user to respond on
71- the timescale of the biological event; for instance, within the integration
72- window of a neuron.
71+ passes between the physical event and the response of the real-time system. A
72+ short latency allows the user to respond on the timescale of the biological
73+ event; for instance, within the integration window of a neuron.
74+
75+ Many acquisition systems rely on a USB connection between the acquisition board
76+ and PC. Or, they rely on closed-source 3rd-party drivers and APIs that are not
77+ optimized for low response latencies. The slower transfer characteristics of
78+ USB means that a typical closed-loop latency is in the range of several
79+ to tens of milliseconds. This is a considerable duration for the brain, as it
80+ is notably longer than the average action potential duration of around 1 ms. On
81+ average, ONIX provides much shorter latencies, of around 150 microseconds, because:
82+
83+ - ONIX is transfers data to the host computer without the CPU via DMA over PCIe.
84+ - ONIX uses a custom device driver optimized for low latency.
85+ - The ONI API allows explicit control over a single parameter to governs the
86+ trade off between data latency and overall bandwidth.
7387
74- Many classic acquisition systems rely on a USB connection between the
75- acquisition board and PC. The slower transfer characteristics of USB means that
76- a typical closed-loop latency would be in the range of several to tens of
77- milliseconds. This is a considerable duration for the brain, as it is notably
78- longer than the average action potential duration of around 1 ms. ONIX has much
79- shorter latencies, of around 150 microseconds, because the host board of ONIX
80- is directly connected to the acquisition PC, in a PCIe slot. This means that
81- the system can respond quickly to detected events. It also means that time is
82- freed up for this detection itself; by reducing the overhead time, more complex
83- analysis can be run to extract the phenomenon you are interested in.
88+ This permits the user to optimize their system's response time for a given
89+ experiment. In all, this means that the system can respond quickly to detected
90+ events. It also means that time is freed up for this detection itself; by
91+ reducing the overhead time, more complex analysis can be run to extract the
92+ phenomenon you are interested in.
0 commit comments