Constraints and advantages of MUSIC vs frequency hopping HMFCW Ranging

[jakapoor]

Reposted from Silva [7/12/2018], #17

The accuracy of subspace techniques (i.e. MUSIC) in determining AoAs is far superior to interfermotery systems that solely use the phase difference of two antennas. This is because subspace techniques are resilient in a non ideal environment with environmental/system inherent interference/noise [3,4,5]. As you suggested in #15, "Work by MacCurdy and Richardson here at Cornell suggested that a completely AOA-based triangulation system will be woefully inadequate to achieve high resolution." I believe the type of system they are referring to is a non-subspace AoA interferometry system. Since frequency hopping would not account for the effects of environmental/system-inherent noise, using a subspace technique with smoothing would be a prerequisite to achieving high accuracy AoA based on previous findings, and not a supplementation.

Reposted from Kapoor [7/12/2018], #17

For this, and other reasons concerning the rigid layout requirements of a 4-antenna system, the decision to use 4 antennas and MUSIC should not be taken lightly.

As you suggested in #15, "Work by MacCurdy and Richardson here at Cornell suggested that a completely AOA-based triangulation system will be woefully inadequate to achieve high resolution." I believe the type of system they are referring to is a non-subspace AoA interferometry system.

This is incorrect. Gabrielson et al. say the following about the AOA system they tested:

Digitally steered phased - array approaches have also been employed to determine tag bearing. These methods use multiple antennas, usually in a circular or linear array, and establish signal direction by measuring the phase difference of the signal at each receiving antenna. This method usually requires multiple synchronized receiver signal paths; modern directional receivers usually accomplish the task with high-speed synchronous analog-to-digital converters (ADCs) operating at the IF stage of the receiver. The sampled multichannel signal is then digitally down-converted and the bearing is established via software; the multiple signal classification (MUSIC) [15] algorithm is widely used for this purpose. The complexity of this approach can be problematic for wildlife tracking applications, and receivers of this type are costly. Additionally, phased-array receive antennas must be mounted far from other objects; experiments conducted by our group showed significant variation in signal phase due to nearby vegetation. This constraint requires phased-array antennas to be mounted on tall, sturdy masts rather than opportunistically placed in trees.

As you can see, they tested out precisely the kind of AOA system we are considering, and found it to be ineffective. This underlies my emphasis on forward compatibility of our system with the HMFCW ranging technique described by Dr. Kan and colleagues.

Since frequency hopping would not account for the effects of environmental/system-inherent noise, using a subspace technique with smoothing would be a prerequisite to achieving high accuracy AoA based on previous findings, and not a supplementation.

The HMFCW (~=frequency hopping) ranging algorithm developed by Dr. Kan and colleagues was designed specifically to deal with the effects of noise, and to keep the system simple and more resilient to complications from layout constraints. They state:

In order to get AoA information from each carrier, different from previous work using low-directivity antenna for indoor locating purposes [10], which resort to building a large M-element antenna array, we simply use 2-element antenna arrays. Although with an increasing size of M, a linear array can be designed with high directivity, and super resolution algorithms such as MUSIC can be applied, a large M will lead to a heavy and cumbersome infrastructure which contradicts our motivation of using miniaturized low-directivity antennas instead of cumbersome high-directivity antennas.

Combining a multi-antenna MUSIC-like algorithm with a multi-frequency approach seems a laudable endeavor, but we need to make sure we don't paint ourselves into a corner by depending on the MUSIC algorithm.

All this said, if we cannot maintain coherence between ground-nodes (without time-consuming recalibration) as we receive multi-frequency signals, we may already have reached a dead-end with forward-compatibility of our system to the ranging approach.

Reposted from Silva [7/13/2018], #17

As you can see, they tested out precisely the kind of AOA system we are considering, and found it to be ineffective. This underlies my emphasis on forward compatibility of our system with the HMFCW ranging technique described by Dr. Kan and colleagues.

Ok, thank you for finding this passage as I had trouble finding this specifically. Is this passage in the direction finding handbook? If so, what page is it on? It is well known now from previous works that the cause of this multipath problem is due to correlated signals within the signal covariance matrix. This passage does not mention the spatial smoothing techniques that I have reiterated for making MUSIC adaptable to multipath environments. I do agree with this passage though on the matter of basestation placement. We have already discussed that we should place our receiver basestations high up in the air (near the top of trees in a forest environment). I do not believe this is a significant constraint as our design calls for a small number of receivers for each several-acre coverage zone. Furthermore, when an open space in the environment is unavailable, the relatively low weight <30 lb basestation (the proposed 5.7 weight APC Back-UPS + 14 lb solar panel will form the bulk of this weight) can be put on a platform. In general, this setup will also be beneficial for the forward compatible system to have less multipath at receivers and greater receiving range.

Combining a multi-antenna MUSIC-like algorithm with a multi-frequency approach seems a laudable endeavor, but we need to make sure we don't paint ourselves into a corner by depending on the MUSIC algorithm. All this said, if we cannot maintain coherence between ground-nodes (without time-consuming recalibration) as we receive multi-frequency signals, we may already have reached a dead-end with forward-compatibility of our system to the ranging approach.

I appreciate the skepticism here. The methodology in the proposal accounts for many of these issues as well as the references provided above (i.e. high antenna constraint and multipath interference - [1,2]). Please consider that we are close to obtaining AoAs from root MUSIC with subspace smoothing (ETA 1-2 weeks away). Then we will reassess.

[jakapoor]

Ok, thank you for finding this passage as I had trouble finding this specifically. Is this passage in the direction finding handbook? If so, what page is it on?

Yes, on p. 1135.

This passage does not mention the spatial smoothing techniques that I have reiterated for making MUSIC adaptable to multipath environments.

Good point. I wasn't aware that spatial smoothing was't an inherent part of the MUSIC algorithm. If Gabrielson et al. didn't try out a MUSIC-based system employing spatial smoothing then we have greater cause to believe the MUSIC+smoothing approach might work better.

I do agree with this passage though on the matter of basestation [i.e. ground-node] placement. We have already discussed that we should place our receiver basestations [i.e. ground-nodes] high up in the air (near the top of trees in a forest environment). I do not believe this is a significant constraint as our design calls for a small number of receivers for each several-acre coverage zone.

Unfortunately, this may be a big problem. We want to avoid costly and difficult to rig masts. Furthermore, in rainforest environments the canopy may be over 30 m high (physically challenging and costly to rig a mast that high), and the spatial scale of activity may be only 30 m in area on the ground (meaning that high ground-nodes will have a highly vertically-angled "view" of transmitters, which will reduce triangulation accuracy). We need to be able to make our system adaptable not only to large open areas, but also very constrained small areas (100 m^2) with dense vegetation. In this latter case the ground-nodes will have to be much closer to the ground, and each other, and in close proximity to tree trunks. This is the primary reason we are anticipating - and trying to correct for - strong multi-path interference in the first place. However, if what you suggest is true (i.e. that Gabrielson et al. didn't use spatial smoothing to test out a phase-based triangulation approach), then we may be able to get by with adventitiously placed ground-nodes.

Please consider that we are close to obtaining AoAs from root MUSIC with subspace smoothing (ETA 1-2 weeks away). Then we will reassess.

Given this assessment, combined with your argument that we're adding subspace smoothing into the equation, I agree we should continue in our current trajectory for now, test the resilience of our system to cluttered environments, then go from there.