Testing the EVB: - 4-bit SDIO interface taken directly from pads on Pandora PCB where wifi module would be - Other digital lines taken directly from pads on Pandora PCB: IRQ, RESETX/PMEN - Main power input (called D3.3V on EVB) comes from battery net on Pandora board (3.3V - 4.2V) - 0R jumper resistors (R15,R16,R26,R28,R30,R41) on EBV where replaced with 33R2 resistors to help control long wire lengths. It did help. - 1.8V LDO regulator on EBV was disconnected (remove R22) and EBV 1.8V supply was provided from 1.8V supply at wifi module footprint on Pandora board. (this connection not shown in picture) - Jumper J3 was removed on EBV to indicate SDIO bus is operated at 1.8V logic. Results (iperf): - Pandora LCD off: 12Mb/s to 13.8Mb/s (only peaked at 13.5Mb/s using EVB s onboard LDO U7, peak was higher using the Pandora s 1.8V supply at 13.8Mb/s) - Pandora LCD on: 11.5Mb/s to 12.5Mb/s - Results were always quite consistent - Web browsing was much faster than with onboard WiFi and Pandora antenna. - The isolation of the chip antenna on the EBV helps to preserve the RF signal.
Pandora rev6a testing: - Rev6a is a test board. There were several changes to the WiFi clocks for testing. 1) Crystal + oscillator driver (same design as U24 but placed right next to WiFi module) 2) 1.8V 26MHz oscillator connected directly to the WiFi module. 3) Original long clock design with some changes: impedance as a stripline controlled to guarantee 50 ohms impedance. The board stackup got changed on rev5a and made the impedance of this trace very low! An older board, rev3 made by a different company had a better impedance but rev6a is the first revision to guarantee it was exactly right. Performance results (LCD Off): 1 and 3 were the same with 11-12.5Mb/s maximum throughput using iperf. 2 was slightly lower but was mainly used back when I was testing with SCP. It was never as good as the others so I didn t solder the jumper to connect it often. Performance results (LCD On): Only 1 and 3 were used during testing. Performance sometimes started out at 9-10Mb/s and quickly dropped, sometimes all the way down to 0.5Mb/s. Data rate consistently dropped with LCD on vs. off.
Other tests on rev6a board: 1) 32.768KHz crystal with oscillator driver didn t work, could not get to oscillate! 2) 32.768KHz oscillator with 1.8V drive output. Results: 1 did not work but 2 worked fine. There was no change in performance.
Further Notes: - PANDORA data rates are best with LCD powered down OR powered up but not connected to LCD cable that goes into lid. This suggests interference is not on main PCB itself. Performance did vary slightly between tests but they always do using iperf. With both boards, the performance would barf at about 20 seconds into a 60 second test, sometimes just after 30 seconds and then would take between 2 to 10 seconds to recover. - Using an external antenna held very still produced consistent results on PANDORA and did not change much at all when LCD was on or off. The biggest inconsistency during this test was my hand moving around while lifting the lid and holding the antenna still. This tends to support the theory that WiFi interference is not leaking through the mainboard itself. The problem is in the lid. - Out of all my clock testing, PANDORA got the best RF performance using the clock design that is on all revisions of the board now (although better impedance control on rev6) or the new clock that was identical to the current one, only placed as a copy right next to the WiFi module. - I believe the source of our troubles is bad antenna design or more specifically, implementation. The antenna design itself with the LCD signals powered down is very close to maximum (within about 1Mb/s as you can read above) and it is possible to try different tuning elements on the PCB to see if I can max that out. Right now, my best attempt gets about 92% of the maximum possible. Figuring out the proper LC tuning elements might lead to getting the other 8% that is missing. - The real problem, as Grazvydas has learned and I only finally can see now that I am using iperf, is that as soon as all that noise from the LCD cable and possible ground loop noise is introduced, the RF signal is destroyed. In fact, on my unit, when I was getting about 0.5Mb/s through iperf, just putting my hand over the entire antenna inside the plastic case upped it to about 4Mb/s. The design is clearly not tuned to the rest of the hardware in the lid environment which was always the hardest thing to do.
Possible source of noise causing RF failure: - A ground loop may exist due to new copper tape providing a secondary path back to the mainboard, but also returning the digital LCD signals as well. - Other complications with antenna placement next to the LCD and some other unknown signal interaction is quite possible too. - I believe this to be the main problem that I need to solve.