I was discussing the idea of a periodized training plan for a cognitive radio with a friend the other day. He had suggested that during low traffic times, such as late at night, it might be a good idea for the cognitive radio to go into high learning mode.
This mode would require more resources which would take away from the main end goal of performance, but because it was a lot priority time the effect on performance wouldn't be as much of an issue. The problem with this logic is that in order for the radio to learn it is going to need inputs in order to develop a history of cases from which to refer to for future learning. During a low traffic period such as late at night there is less traffic in the over all spectrum environment and there for less of a chance to take in inputs from which to develop decisions which in turn build up the case based library.
He had a great idea to create an artificial stimulus which could create inputs for the system. These inputs would simultate things such as other radios operating in the same spectrum. The cogntive node would not know that this is actually an artificial stimulus and would follow along its learning path of reading the meters, making a decision based on a policy tree and case based reasoning flow chart. These decisions contribute to the overall case library.
This learning stimulus would engage during low period times which is an efficient use of the radio. While it might typically lay dormant during these times of low traffic with the use of the LSM it can train itself without affecting the end goal of overall performance. The LSM can shut down when the radio needs to send priority information, or when there is real spectrum stimulus in the area.
Saturday, March 1, 2008
Wednesday, February 27, 2008
Soar Cognitive Engine
There was a paper included in this week's class readings that discussed Applications of Machine Learning to Cogntive Radio
It was an interesting paper and dealt mainly with fundamental ideas about learning, reasoning as they can be applied to cognitive radio. However at the very end of the paper they mentioned a real implementation of the ideas by using an Ossie Software defined radio and the Soar Cognitive Radio Engine.
There is a little bit more detail of their implementation here.
I am in the process of learning more about this Soar Cognitive engine. It is a general computational model that is geared towards a Unified Theory of Cognition. Not that I know what that means yet. It seems to have found application in many different fields from psychology to AI in video games and most notably for me Cognitive radio.
This could be an a good avenue to explore for research given the little bit of work using it in a cognitive radio framework?
I saw that an MS student did a thesis on using the OSSIE SCA to develop a software defined 802.16 system.
I wonder if this implementation could be adapted with the Soar cognitive engine to develop a cognitive WiMAX system? It looks like it's doable since the above papers have followed a similar approach of synergy between OSSIE and Soar. It would still be novel because it is a WiMAX implementation.
It was an interesting paper and dealt mainly with fundamental ideas about learning, reasoning as they can be applied to cognitive radio. However at the very end of the paper they mentioned a real implementation of the ideas by using an Ossie Software defined radio and the Soar Cognitive Radio Engine.
There is a little bit more detail of their implementation here.
I am in the process of learning more about this Soar Cognitive engine. It is a general computational model that is geared towards a Unified Theory of Cognition. Not that I know what that means yet. It seems to have found application in many different fields from psychology to AI in video games and most notably for me Cognitive radio.
This could be an a good avenue to explore for research given the little bit of work using it in a cognitive radio framework?
I saw that an MS student did a thesis on using the OSSIE SCA to develop a software defined 802.16 system.
I wonder if this implementation could be adapted with the Soar cognitive engine to develop a cognitive WiMAX system? It looks like it's doable since the above papers have followed a similar approach of synergy between OSSIE and Soar. It would still be novel because it is a WiMAX implementation.
Tuesday, February 26, 2008
Just some initial thoughts on research project
Cognitive WiMAX
Justifications
Enable cost effective deployments
Minimize need for expensive System Integrators
Enable less skilled labor to install
Self starting / Autonomous initialization
Important for DOTs and large scale deployments
Important for rural and developing/3rd world nations
Fast deployments for emergency and hostile environments
Account for changes and crowding of spectrum
Cost effective long term operations
Minimize requirement for in-field service
Capability to adapt to changing wireless landscape
Justifications
SDR based radios vs System on Chip
Higher return on initial investment
Lifespan of 15-20 years instead of 3-5 years based on how fast technology improves
Lower cost installation and maintenance
Synergy with the way a DOT or government does business
One time high dollar amount procurements w/o budget for upgrade/maintenance
Improved Performance
Make better use of available spectrum
Optimize configuration based on existing conditions improve overall throughput/performance
General Concept
Some Ideas
Spectrum Decision Methodology
User requirements System knobs
Data rate
Acceptable error rate
Delay bound
Transmission mode
Bandwidth requirements
Carrier frequency
Adapting WiMAX transport protocols to a dynamic spectrum environment
Ideas (cont)
Handover Delay
Delay caused by spectrum handoff
Contrast to delay caused by mobile handoff
Synergy with my current work with Cisco quantifying handoff delay between APs from a mobile client
Co-existence of WiMAX with other WiMAX and 802.11 in unlicensed bands
How the scheduled MAC of WiMAX interoperates with a CSMA/CD based systems like 802.11
Adapting Sports Physiology Models to ‘training’ a CR
In Sports Physiology there is a concept known as Periodized Training
Variables such as Training volume and Training intensity are manipulated throughout the year in order to optimize performance at a specific time such as for the competition phase or a specific event as the finals
Research
Develop a ‘training’ plan for a CR radio node
Adapt physiological models of periodized training to CR
Develop schedule for ‘when’ a CR should be in a certain mode of operation in order to optimize the performance during certain events.
Similar to taking an athlete and training them for a competition season or a specific event
Matveyev’s Model of Periodization
For example
In the prepatory phase, a CR would be in a spectrum sensing mode
Power usage HIGH
Cooperative with other nodes higher overhead
LOW throughput lower performance
In the ‘competition’ or high performance phase
Lower sensing , lower cooperation, more resources for throughput higher performance
The CR knowingly sacrifices performance during non-mission critical times in order to learn more which in the end should optimize the end goals when needed for higher priority data
Thursday, February 7, 2008
definitions
Lots and lots of definitions of cognitiv radio.
I like James Neel's definition from this presentation:
A cognitive radio is a radio whose control processes
permit the radio to leverage situational knowledge
and intelligent processing to autonomously adapt
towards some goal.
I like James Neel's definition from this presentation:
A cognitive radio is a radio whose control processes
permit the radio to leverage situational knowledge
and intelligent processing to autonomously adapt
towards some goal.
Cognitive WiMAX vs typical cognitive radio
Most of the emphasis on cognitive radio is for a secondary user to operate in the same spectrum as a primary user. So there is a lot of focus on dynamic spectrum access and the ability to sense the presence (and modulation code) of a primary user and then vacate the channel.
But with WiMAX MAC is built around a scheduler. Where a base stations schedules packest to users so there is no need for carrier sense and collision detection. The goal was a higher quality of service. But with this dedicated nature of base station to subscriber there is not as much emphasis on searching for a primary user, because teh WiMAX is the primary user.
There is more opportunity to use cogntive abilities to optimize the quality and throughput between endpoints. For example, just like 802.11 based systems, WiMAX will scale up or down the bandwidth depending on how far away the receiving radio is. Closer together--> higher signal--> controller sets a higher bandwidth modulation.
"system would use cognitive radio technology to identify interference and poor links and then change its own signal transmission to improve the weak links." taken from here
So the only factor here is proximity. If more information was used to make the decision such as Bit error rate, snr... a better decision can be made.
A more driving force behind a cogntive WiMAX would be self initialization for easy deployment/startup and long term operations for emergency and 3rd world countries.
There is some effor in 802.16h for interoperation in unlicesened bands but that is different than a complete cognitive function.
But with WiMAX MAC is built around a scheduler. Where a base stations schedules packest to users so there is no need for carrier sense and collision detection. The goal was a higher quality of service. But with this dedicated nature of base station to subscriber there is not as much emphasis on searching for a primary user, because teh WiMAX is the primary user.
There is more opportunity to use cogntive abilities to optimize the quality and throughput between endpoints. For example, just like 802.11 based systems, WiMAX will scale up or down the bandwidth depending on how far away the receiving radio is. Closer together--> higher signal--> controller sets a higher bandwidth modulation.
"system would use cognitive radio technology to identify interference and poor links and then change its own signal transmission to improve the weak links." taken from here
So the only factor here is proximity. If more information was used to make the decision such as Bit error rate, snr... a better decision can be made.
A more driving force behind a cogntive WiMAX would be self initialization for easy deployment/startup and long term operations for emergency and 3rd world countries.
There is some effor in 802.16h for interoperation in unlicesened bands but that is different than a complete cognitive function.
Random cogntive radio research ideas
I read a great paper that outlined several potential research issues with cognitive radios:
[1] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. MOhanty, \NeXt generation/dynamic access/cognitive radio wireless networks: A survey," Elsevier Computer Networks Journal, vol. 50, pp. 2127{2159, September 2006.
1) Latency caused by spectrum handoff
-In the work I've been doing lately with mobile roaming we have been quantifying the latency delay created during a handover between a mobile node and a roadside accesspoint. This delay can be instrumental in causing issues with applications such as video.
A research idea would focus on the latency delay that would be incurred with a cogntive radio when it decides to switch spectrums. Just like with a layer 2 handoff there is going to be some latency when the radio and it's partner or network decides to vacate spectrum and move to a new allocation.
2)Identifying spetrum bands based on combining many characteristics and not just if the spectrum is available or not
-capacity (which is affected by interference level and path loss)
-delay
-link error rate
-holding time
==> selecting the appropriate spectrum for the particular application
3)Spectrum decision.
What are the factors that go into makin a decision in which spectrum to switch to.
-User requirements drive this decision
-data rate
-acceptable error rate
-delay bound
-transmission mode
-bandwidth requirements
4) routing algorithms in an open spectrum enviroment.
-Routing has never had to think about open spectrum environments This adds a new dimension to routing protocols and how they are inter related to spectrum management
5) Transport protocols for xG networks
TCP/UDP for dynamic specturm environments and cooperation with upper and lower layers. Again same as above issue, current transport protocols have not had to consider this issue in the past.
6) Spectrum sensingin OFDM based networks. OFDM seems to be the darling for cogntive radio applications. And sensing i guess can be performed quickly but it is complex, but because there are a large # of carriers the sensing algorithms need to be optimized so the # of samples needed to detect primary users is minimzed within a given detection error probablity.
7) just a conceptual idea about combining Remote Wireless Sensor networks with a cognitive radio network. Sort of pie in the sky and maybe not viable. By nature a RWSN is going to be cheap and easily deployable, low power. stand alone. In contrast a cogntive radio is more complex. Can you tie the two together. What about building a very cheap energy detetctor for a particular block of spectrum. In fact build 10 different ones to cover a range of spectrum and put them on their own RWSN mote. Deploy them around a CR node. Let the motes perform the sensing and send that information to the CR node.
[1] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. MOhanty, \NeXt generation/dynamic access/cognitive radio wireless networks: A survey," Elsevier Computer Networks Journal, vol. 50, pp. 2127{2159, September 2006.
1) Latency caused by spectrum handoff
-In the work I've been doing lately with mobile roaming we have been quantifying the latency delay created during a handover between a mobile node and a roadside accesspoint. This delay can be instrumental in causing issues with applications such as video.
A research idea would focus on the latency delay that would be incurred with a cogntive radio when it decides to switch spectrums. Just like with a layer 2 handoff there is going to be some latency when the radio and it's partner or network decides to vacate spectrum and move to a new allocation.
2)Identifying spetrum bands based on combining many characteristics and not just if the spectrum is available or not
-capacity (which is affected by interference level and path loss)
-delay
-link error rate
-holding time
==> selecting the appropriate spectrum for the particular application
3)Spectrum decision.
What are the factors that go into makin a decision in which spectrum to switch to.
-User requirements drive this decision
-data rate
-acceptable error rate
-delay bound
-transmission mode
-bandwidth requirements
4) routing algorithms in an open spectrum enviroment.
-Routing has never had to think about open spectrum environments This adds a new dimension to routing protocols and how they are inter related to spectrum management
5) Transport protocols for xG networks
TCP/UDP for dynamic specturm environments and cooperation with upper and lower layers. Again same as above issue, current transport protocols have not had to consider this issue in the past.
6) Spectrum sensingin OFDM based networks. OFDM seems to be the darling for cogntive radio applications. And sensing i guess can be performed quickly but it is complex, but because there are a large # of carriers the sensing algorithms need to be optimized so the # of samples needed to detect primary users is minimzed within a given detection error probablity.
7) just a conceptual idea about combining Remote Wireless Sensor networks with a cognitive radio network. Sort of pie in the sky and maybe not viable. By nature a RWSN is going to be cheap and easily deployable, low power. stand alone. In contrast a cogntive radio is more complex. Can you tie the two together. What about building a very cheap energy detetctor for a particular block of spectrum. In fact build 10 different ones to cover a range of spectrum and put them on their own RWSN mote. Deploy them around a CR node. Let the motes perform the sensing and send that information to the CR node.
Saturday, January 19, 2008
Rapid Deployment and Maintenance Free operation of Wireless infrastructure
This is a potential idea for a dissertation. One of the biggest cost factors for a large scale deployment of a wireless system isn't really the hardware itself. The hard infrastructure (cabinets, poles, grounding, pads) etc. costs a lot as well. But I think the biggest cost is the systems integrator costs for design, installation, and long term maintenace.
The design and deployment is complicated enough that you'd need a batch of highly skilled technical people to get them out there. One unforseen change in spectrum usage from a random noise signal and it will have to be changed. Again, call out the Masters educated techs and fix it.
A cognitive radio based wireless infrastructure would be able to adapt to changes in the wireless environment, and be easily deployed. It wouldn't take a highly skilled person to tack them up on a pole and run wires to them.
The DOT is in a position right now where their maintenance forces are severely taxed just for the basic infrastructure such as signals. Their workforce is untrained in basic IP based devices such as cameras and wireless radios. So in order to maintain it themseleves they will have to 1) educate their workforce and 2) increase their workforce. Right now it seems that they don't want to do either and are even going in the opposite direction in terms of downscaling and privatizing maintenance functions.
but that privitazation is going to increases costs significantly especially when you have a technical product and complex function as a wirelss infrastructre.
A cognitive based system would be require a variet of PHY, MAC and routing options This might be a good dissertation topic. In addition to a cognitive radio aspect I think there is a large need for a cogntive network aspect. In the MacKenzie, Da Silva paper on cognitive networks they mention
"For instance a particular MAC protocol may optimize for power consumption, creating higher hop count routes that use short links. Howeer this mode of operation might result in additional end-to-end delay (due to the additional processing, queing and transmission delay that goes along with higher hop count routes) which in turn could affect the transport layer leading to more transmissions
I've seen first hand how higher hop counts lead to decreased throughput. So this issue is very important. A single cognitive radio might have it's own selfish requirements that will help it's overall goal but an overriding umbrella of a cognitive network with emphasis on a systems end-to-end goal is going to take each radio's decisions into account and make the best overall change to the system operation.
The design and deployment is complicated enough that you'd need a batch of highly skilled technical people to get them out there. One unforseen change in spectrum usage from a random noise signal and it will have to be changed. Again, call out the Masters educated techs and fix it.
A cognitive radio based wireless infrastructure would be able to adapt to changes in the wireless environment, and be easily deployed. It wouldn't take a highly skilled person to tack them up on a pole and run wires to them.
The DOT is in a position right now where their maintenance forces are severely taxed just for the basic infrastructure such as signals. Their workforce is untrained in basic IP based devices such as cameras and wireless radios. So in order to maintain it themseleves they will have to 1) educate their workforce and 2) increase their workforce. Right now it seems that they don't want to do either and are even going in the opposite direction in terms of downscaling and privatizing maintenance functions.
but that privitazation is going to increases costs significantly especially when you have a technical product and complex function as a wirelss infrastructre.
A cognitive based system would be require a variet of PHY, MAC and routing options This might be a good dissertation topic. In addition to a cognitive radio aspect I think there is a large need for a cogntive network aspect. In the MacKenzie, Da Silva paper on cognitive networks they mention
"For instance a particular MAC protocol may optimize for power consumption, creating higher hop count routes that use short links. Howeer this mode of operation might result in additional end-to-end delay (due to the additional processing, queing and transmission delay that goes along with higher hop count routes) which in turn could affect the transport layer leading to more transmissions
I've seen first hand how higher hop counts lead to decreased throughput. So this issue is very important. A single cognitive radio might have it's own selfish requirements that will help it's overall goal but an overriding umbrella of a cognitive network with emphasis on a systems end-to-end goal is going to take each radio's decisions into account and make the best overall change to the system operation.
Subscribe to:
Posts (Atom)