July 26, 2021
Removing the effect of speckle from LiDAR pointclouds
Back to the Inner Circle

Next generation LiDAR sensors are inching closer to L4 ready LiDAR technology by trading in limiting direct detection methods for homodyne detection. This detection method differs from legacy LiDAR technology in that it is not only capable of measuring the amplitude (or intensity) of the returning light, but it is also now sensitive to other optical properties such as frequency and phase. The downside? Many traditional FMCW LiDAR technologies that are now leveraging homodyne detection are unable to solve pointcloud blur or speckle.

Measuring these new properties of the returning light signal is what enables the direct measurement of doppler per point, which significantly aids perception algorithms in object detection, classification and tracking. On top of this, homodyne detection technology is inherently more sensitive, meaning that it can pick up weaker signals from further away.

In other words, homodyne detection allows LiDAR to extract more information from the signal, enables per-point doppler measurements, and increases its detectability of objects at longer ranges. For these reasons, there is tremendous value and market drivers for making this technological leap.

However, despite the clear advantages of homodyne detection, it often comes with a very significant disadvantage: speckle. Baraja’s Spectrum-Scan™ LiDAR platform is one of the few market options that has successfully harnessed all of the benefits of homodyne without any of the tradeoffs, like speckle.

What is speckle?

Speckle is an inherent issue with any form of detection that is sensitive to the phase of the light, as is the case in homodyne detection. In generalized terms, speckle is due to the sum of all the phase orientations in the returning light adding to zero, which results in missed detections and therefore missing points within the pointcloud, creating a “speckled” or grainy image that lacks necessary detail and accuracy.

When a well-collimated beam of light, which is an attribute of a good LiDAR sensor, hits a typical surface, the material properties of that surface generate different phases on different parts of that beam. To measure this returning signal, the LiDAR needs to effectively calculate and measure the sum of all of these phase orientations. Sometimes, these phases add up nicely and lead to a strong return signal. However, more often than not, the phases sum to a near zero value, meaning you are unlikely to receive a return at all.

This effect happens to every point in the pointcloud, so over an entire frame, the output pointcloud appears to be speckled or grainy and lacks a lot of the detail that would have otherwise been in the pointcloud, as indicated in the photo below.

This file is licensed under the Creative Commons.

As you can see in the photo, if there is too much speckle in the output data, it’s as if the system has drastically reduced the points per second and, therefore, reduced the effective points on the target. This makes it difficult for the LiDAR to accurately and precisely measure what the actual object is and determine things such as the size, shape and velocity of the objects at long range within the field-of-view (FoV).

To make matters worse, speckle is unpredictable. Because of the inherently random nature of this phase summation for each point in the FoV, it’s impossible to know beforehand whether or not a point will be affected and go undetected. Even for very strong returns that are otherwise highly detectable objects, such as walls and vehicles, there is still a strong possibility of having returns that equate to zero. Ultimately, this phenomenon significantly impacts the effective range to which these objects can be segmented and classified in the pointcloud.

Overall, speckle drastically reduces a LiDAR sensor’s probability of detection (PD), which is one of the key parameters for long-range detection and what is primarily used to measure the sensitivity of LiDAR. The lower your PD, the less likely you are to get a return.

For example, if your LiDAR sensor had enough resolution to get a potential of 10 returns on an object at a given range and you had 100% PD, you would get all 10 points back, giving you a clear image of what’s ahead. If instead, you only had a 10% PD at that range, you would only receive one point back on average, which isn’t enough to classify or segment an object. This example shows that with a reduced PD, it’s much harder to use the sensor for long-range applications.

How speckle creates “trust” issues

The problem long-range LiDAR is trying to solve is enabling highway driving at highway speeds. To do this, you need a high enough resolution to get enough potential points on a target and enough sensitivity to really make use of that resolution and enable the detection of small, dark objects.

Given that in order to do this, you need to push the boundaries on all fronts of the technology, meaning you are already maximizing resolution and sensitivity. Speckle effectively destroys LiDAR’s maximum effective range and serves as a fundamental blocker to highway driving capability. Without reliable pointcloud data and clear details, a LiDAR sensor cannot effectively detect and record which objects are up ahead, let alone how large they are, if they’re moving, or other important details.

At Baraja, we believe true L4 technology starts with data you can trust. If speckle is not solved, then the data presented by that LiDAR is too unreliable to be used for L3 or L4 features.

Solving the issues of speckle

To solve the extremely challenging problem of enabling long-range, high speed autonomous driving, LiDAR technology is already pushing the physical and technological limits to achieve the range, resolution and cost required. Therefore, simply increasing the points per second (PPS) of the system is not an option. Instead, LiDAR manufacturers must solve the fundamental issue of speckle.

At Baraja, we have eliminated the issue of speckle from our LiDAR sensor, making our LiDAR that much more ready for L3 and L4 autonomous vehicles.

Using a custom integrated photonic receiver, we are able to extract the previously missed phase interference information, eliminating speckle from the product. With this breakthrough technology, we are able to fill in the previously missing gaps, or speckles, in the pointcloud and ensure we have the best-in-class effective range that is ready for highway driving autonomous features.

Producing a speckle-free pointcloud is just one of the many next generation LiDAR problems our product has solved. Paired with our unmatched resolution, range and reliability, the Spectrum-Scan™ sensor is one of the first and only LiDAR sensors on the market capable of meeting the strict demands of L4 autonomous driving.

If you’re searching for an endgame LiDAR, contact our team today.

Latest Articles
November 29, 2023
Which chip are they talking about? The power of integration: new-to-world Doppler RMCW Spectrum-Scan™ LiDAR
Federico Collarte
Baraja announces the availability of A-Samples of all the integrated components necessary to build a new-to-world Doppler RMCW Spectrum-Scan™ LiDAR for automotive integration.
Learn more
January 5, 2023
Press Release: Baraja and TIER IV strengthen collaboration with signing of memorandum of understanding ahead of CES 2023 exhibit
The Baraja Team
Baraja, creator of the breakthrough Spectrum-Scan™ LiDAR technology, is expanding upon its work with TIER IV, an open-source autonomous driving leader, by signing a memorandum of understanding (MOU) to jointly validate, develop and optimize their respective autonomous driving solutions.
Learn more
Join the Baraja Inner Circle
Subscribe to receive the latest from Baraja