433 MHz LoRa SAW Bandpass Filter

 The LoRa network that a team of radio amateurs in Bucharest is testing is - for various reasons - deployed in the 433 MHz ISM band.
A check of the radio spectrum showed that in this band there is a rather high level of noise generated by the multitude of competing networks in the 380 - 500 MHz band, most of them transmitting with high power.
These contribute to an increased noise level which in turn affects the signal to noise ratio of the receiver (usually SX1273).

 According to the SX127x datasheet:

"SX1272/73 Is a half-duplex, low-IF transceiver. Here the received RF signal is first amplified by the LNA. The LNA input is single ended to minimize the external BoM and for ease of design. Following the LNA output, the conversion to differential is made to improve the second order linearity and harmonic rejection. The signal is then down-converted to in-phase and quadrature (I&Q) components at the intermediate frequency (IF) by the mixer stage. A pair of sigma delta ADCs then perform data conversion, with all subsequent signal processing and demodulation performed in the digital domain. The digital state machine also controls the automatic frequency correction (AFC), received signal strength indicator (RSSI) and automatic gain control (AGC). It also features the higher-level packet and protocol level functionality of the top level sequencer (TLS)".

Although the circuit has been designed with interference rejection (IMD) in mind and the modulation itself is designed to ensure communication in environments with radio-frequency pollution and high signal-to-noise ratios, any noise reduction before being processed in the input stages is beneficial.

Filtering signals outside the operating frequency range is, however, a costly process in which a number of compromises have to be adopted.

Although the circuit has been designed with interference rejection (IMD) in mind and the modulation itself is designed to ensure communication in environments with radio-frequency pollution and high signal-to-noise ratios, any noise reduction before being processed in the input stages is beneficial.

Filtering signals outside the operating frequency range is, however, a costly process in which a number of compromises have to be adopted.
Their advantage is the narrow passband and the disadvantage is the insertion attenuation.
There are a number of such filters available for purchase from manufacturers in China but the overwhelming majority of them are realized with 40 MHz bandwidth (@-3db) filters, which allow signals in the 436-440 frequency range, where the output signals of amateur radio repeaters are found, to pass at increased levels.
At least in Bucharest, one of them is in DMR system and emits with increased power of several tens of W, which has the ability to contribute significantly to S/N degradation.

 

 

Another issue noticed was a poor PCB design on which these SAW filters are installed to become a usable product in the shack.

These filters must also ensure a considerable rejection of the transmitting stations of mobile phone network transmitters in the 800 MHz - 1 GHz band.
A look at the way the wiring on which the SAW filter is placed suggests a number of design problems that degrade its performance.
Because I'm a speedy ham, I thought I'd make my own filter, with much better performance than those available at online stores.

So, I started reviewing the range of SAW filters available from reputable manufacturers and settled on Qualcomm.

And after a "painful" and lengthy selection process I settled on a number of three filters that could be a much better solution than the existing ones, for the purpose I proposed.

One of them is  B39431B3735H110.

Now, on the PCB, the manufacturer is giving us some hints about how the PCB must be designed to get the most of what the manufacturer's specs:

"Minimising the crosstalk
For a good ultimate rejection a low crosstalk is necessary. Low crosstalk can be realised with a good
RF layout. The major crosstalk mechanism is caused by the “ground-loop” problem.
Grounding loops are created if input-and output transducer GND are connected on the top-side of
the PCB and fed to the system grounding plane by a common via hole. To avoid the common
ground path, the ground pin of the input- and output transducer are fed to the system ground plane
(bottom PCB plane) by their own via hole. The transducers’ grounding pins should be isolated from
the upper grounding plane.
A common GND inductivity of 0.5 nH degrades the ultimate rejection (crosstalk) by 20 dB.
The optimised PCB layout, including matching network for transformation to 50 Ohm, is shown
here. In this PCB layout the grounding loops are minimised to realise good ultimate rejection".

 

Just compare this with the footprint of the cheap Chinese filters shown above...

And here they are, assembled:

And here are the masurements, taken with SatSaGen with Adalm Pluto as VNA with tracking generator: