For the past few months, I’ve been working on some interesting challenges with our latest generation of hardware. The Sense monitor collects detailed, sub-second level data that allows us to help you understand your electricity usage. It is a complicated piece of equipment that relies on many different subsystems working in harmony. I thought it would be interesting to illustrate one of the many challenges involved in building hardware designed to live inside the home’s electric panel.
The monitor is composed of two separate printed circuit boards (PCBs). One, the power board, handles the dangerous voltages inherent in a product that resides in an electrical panel. It also houses the switching power supply and some of the initial signal processing circuitry. The second PCB, the processor board, does all the heavy lifting. This is where the processor disaggregates the power your home uses so that Sense can give you device by device information. This board also includes the WiFi/Bluetooth module.
The WiFi turned out to be an interesting challenge. Safety is of paramount importance here at Sense. When considering design, we always have to keep in mind Sense’s location inside the electrical panel. In order to obtain the necessary safety certifications, anything protruding from the panel (like Sense’s antenna) must be galvanically isolated. In the first generation of the Sense monitor, we met this requirement by using several digital isolators and an isolated power supply. This methodology worked, but was suboptimal for at least two reasons. First, the digital isolators were expensive. When manufacturing any type of hardware, Bill of Material (BOM) costs can never be too low. A $2 savings over millions of units adds up quickly! Secondly, they caused headaches during the unintentional radiated emissions portion of the FCC EMI certification process.
In redesigning the WiFi isolation solution, at least six requirements needed to be addressed. The new isolation topology had to:
1) Provide acceptable galvanic isolation.
2) Reduce the BOM cost.
3) Maintain or improve EMI performance.
4) Fit into a constrained area as determined by the enclosure.
5) Be designed and debugged under tight time constraints.
6) Maintain WiFi performance.
After exploring many possible design topologies, the option that won out was a completely passive tapped-line input hairpin bandpass filter (BPF).
A hairpin BPF is a type of distributed element (in contrast to lumped element) filter that is commonly used at higher frequencies. As signal wavelengths approach the physical distance they are traveling, lumped element models become unreliable and transmission line theory takes over. Transmission line models allow us to obtain practical results without resorting to Maxwell’s equations.
The BPF is constructed using parallel-coupled microstrip transmission lines fabricated directly on the PCB. Microstrip, as opposed to other types of transmission lines, was chosen due to its ease of fabrication and favorable coupling characteristics. We used a Butterworth filter with a center frequency of 2.445 GHz (the center of the 2.4 GHz WiFi frequency band) and a bandwidth of 90 MHz in order to optimize the filter characteristics while allowing for space and isolation constraints. We were limited to a 2nd order Chebyshev because each higher order adds an additional “hairpin” to the design and PCB area was at a premium. After settling on a 2nd order BPF, we set the characteristic impedance to be 50Ω in order to match the WiFi module and the antenna and we set the spacing between the two “hairpins” to be 1.2 mm, which is the spacing required to pass the isolation safety requirements. From there it was a matter of optimizing the forward gain, S21, and the reflection coefficient, S11. This is done by simultaneously adjusting the width, length, and spacing of each leg of the filter. We wanted to ensure that S21 was as flat as possible across the frequency band while simultaneously minimizing S11 (which maximizes power transfer). Below, you can see the optimized layout and the fabricated PCB.
As soon as we received the new PCBs back from the assembly house, we went into the lab to compare the real world results with the simulations. As you can see below, the real world results very closely match the simulation results. We are able to obtain greater than 12 dB of S11 (VSWR ≤ 1.7) and about 7 dB of S21 variation across the entire frequency band. One thing to note is that the 7 dB of S21 variation can be almost completely compensated out by slightly increasing the output power of the WiFi module.
So let’s take a look at our design goals from before and see what we have accomplished.
We are always working hard to provide you with the best possible hardware and the most engaging experience. I hope this article gives a little bit of insight into just how we go about solving some difficult engineering problems to bring you the quality data you expect, and as always, Happy Sensing.