What could be the main reason for the sensitivity loss in Direct-Tie configuration with Si446x?
Considerable sensitivity loss can be observed in Direct-Tie board configuration (i.e. TX and RX paths are directly connected to each other without the use of external RF switch) with Si446x even if the transmit spectrum looks okay and the 4-element balun matching network is also good in terms of layout and component values (i.e. good RX sensitivity in Split board configuration). Typical problem what can cause even more than 10 dB sensitivity loss is the L0-C0-CPAoff resonance what shunts the RX path in Direct-Tie configuration in RX mode.
In Split configuration (i.e. separated TX and RX paths) optimum TX and RX performances can be achieved. However, in Direct-Tie configuration the TX path (typically L0-C0) needs to be slightly de-tuned (from the optimum TX split values) in order to avoid the above mentioned resonance what would shunt the RX path in RX mode. This is a typical effect in the high frequency bands (i.e. 868-915 MHz). The recommended way of the matching network tuning for the Direct-Tie topology is to get the resonance away from the RF carrier with the tuning of L0 value and then tune the C0 value to get back an acceptable TX performance (tuning of L0 has the significant effect on the L0-C0-CPAoff resonant frequency, since the CPAoff is a fixed about 1.5 - 2 pF, Silicon Labs recommend to tune this resonance to the higher frequencies, i.e. reduce the L0 value, and then slightly increase the value of C0). A few nH in the L0 value can cause considerable effect on the RX sensitivity, so it is also highly suggested copying the RF layout from Silicon Labs reference designs and place the L0, C0 components (with the connection point of RX to the TX path) as close the TX pin of Si446x as possible. A few mm longer trace between the TX pin, L0, C0 and Direct-Tie connection point (compared to Silicon Labs reference designs) introduces a few extra nH what is added to L0 and it can bring the L0-C0-CPAoff resonance back to the RF carrier.
Please refer to the "5.4. Detailed Matching Procedure for Direct Tie Board Configuration" section in AN627 application note for a more detailed explanation and simulations.
How can I further reduce the output power level of Si4010?
The guaranteed output power level range of the Si4010 RF transmitter IC is between +10 and -13 dBm. This power output range can be covered by the settings of PA_Power_Level (bLevel) and PA_Max_Drive (bMaxDrv) API registers included in the vPA_Setup() function. The minimum, -13 dBm, output power level can be achieved with bLevel=0 and bMaxDrv=0.
However, the Si4010 RF chip is capable to transmit with a lower power output than -13 dBm. For this, the PA_LVL SFR has to be directly written. The address of the PA_LVL SFR is 0xCE. For a more detailed register description please refer to the "SFR Definition 12.1. PA_LVL" section in the Si4010 datasheet.
After the vPA_Setup() function in the code the PA slice and bias settings can be overwritten by writing the PA_LVL SFR. For reference, the bLevel=0 and bMaxDrv=0 API settings are equivalent with the PA_LVL SFR value of 0x32. Smaller value of this SFR will provide lower power output than -13 dBm.
Can I have an access to the S-parameters for the LNA and PA pins of Si446x?
Silicon Labs provide impedance parameters (i.e., can be converted into S-parameters) for the RX pins (differential LNA pins) documented in the AN643 application note.
However, Silicon Labs do not supply S-parameters for the TX pin, since the Si446x device works in switched mode (switched-PA mode such as Class-E or Square-Wave, or switched-current mode) and the matching methodology is not the conventional complex conjugate one. Please refer to the related application notes here such as AN627 and AN648.
Which data rate, deviation and modulation index should I use?
Based on http://community.silabs.com/t5/Wireless-Knowledge-Base/Modulation-choice/ta-p/144373, the recommended modulation is 2GFSK, so from now on, let's consider only this type of modulation.
The occupied bandwidth of a 2GFSK signal can be estimated based on the following formula:
The modulation index can be calculated in the following way:
According to the formula for occupied bandwidth, the OBW value is higher with higher data rate and deviation. If one is using lower OBW, the receiver BW can also be set lower, which reduces the noise level in the receiver, and this way improves sensitivity. Hence it seems a good idea to transmit with as low data rate as possible, but applications often require a minimum data rate to send the necessary information in the available time slot. The data rate and the crystal tolerance should be chosen to stay within the allowed channel under extreme conditions as well according to the related regulation standard. Beside this, depending on the crystal accuracy the possibility of frequency offset between the transmitter and the receiver can grow. The right choice for data rate and crystal type depends on several parameters: carrier frequency, carrier power, channel bandwidth, modulation index, and of course the data rate and the crystal type cannot be considered separately. A higher frequency signal introduces higher absolute frequency error with the same type of crystal. A high power application can violate the modulation bandwidth regulation easier, especially for narrow band channels (e.g. 25 kHz). Also, the modulation index determines the occupied bandwidth: higher modulation index introduces higher bandwidth.
To investigate the modulation index, let’s consider 3 typical instances: H = 1, H = 0.5, H = 2. If we choose the desired data rate for our application, the deviation determines the used modulation index. Also, the deviation determines the distance between the transmitted symbols: higher modulation index introduces more distance between symbols. Simultaneously, a higher data rate introduces higher occupied bandwidth, which causes that wider bandwidth is necessary in the receiver, this way a higher thermal noise will appear and the sensitivity will be worse. Thus, by increasing the deviation, the distance between symbols will also grow, which improves sensitivity, but on the other side, the occupied bandwidth will be wider, which causes sensitivity loss. These two effects seems to compensate each other, but there is an optimal choice, at H = 1. This is the value where the receiver sensitivity is at the optimal level, thus this is the recommended modulation index.
To sum up, the recommended modulation index for 2GFSK modulated signals is H =1. The data rate and the crystal type should be chosen based on the requirements of the application to meet the standard regulations. If the data rate is chosen, the H =1 modulation index determines the necessary deviation.
Which modulation (OOK, 2FSK, 2GFSK, 4FSK, 4GFSK) should I use?
Let’s compare the above listed modulation types in pairs.
1. OOK vs. FSK
OOK is a digital amplitude modulation type, which represents 2 different states: the presence or absence of the carrier based on the symbol to be transmitted. Contrarily, for FSK modulation the amplitude of the transmitted signal is fixed, but the frequency can be different based on the given symbol.
The consecutive on-off switching of the PA for OOK modulation causes that the bandwidth increases compared to FSK modulation using the same data rate, this way FSK is a more spectrally efficient modulation. The wider bandwidth makes it necessary to apply a wider filter in the receiver, which increases the noise level and this way the sensitivity will get worse. If the occupied bandwidth is the same for OOK and FSK modulation, the sensitivity is similar (but in this case OOK modulation has lower data rate to have the same bandwidth as FSK). Beside these, OOK modulation is more sensitive for fading, since the information is carried only by the amplitude. Along with this, FSK modulation is more sensitive for the frequency offsets between the transmitter and the receiver.
To sum up, FSK modulation is more spectrally efficient so has better sensitivity and is less sensitive for fading. OOK modulation is less sensitive for the frequency inaccuracy, and thus is commonly used in applications where the frequency accuracy can not be guaranteed.
The above figure shows the spectrum of an OOK modulated signal with 40 kbps data rate and a 2FSK modulated signal with 40 kbps data rate and 20 kHz deviation.
2. FSK vs. GFSK
FSK modulation creates high level spurious contents (at integer multiples of the symbol rate) as well as relatively high side lobes on the transmitter side, which can cause regulation standard violations. A Gaussian filter can be applied to the symbols before creating the frequency modulated signal to suppress these spurs and side lobes by smoothing the baseband signal. This way the bandwidth can be slightly reduced, but the distance between symbols will decrease which causes a slightly worse receiver sensitivity (~0.5 dB) for GFSK modulated signals. Still, GFSK is a generally used modulation type, since it reduces spurious contents on the transmitter side significantly and the loss in sensitivity is negligible.
The above figure shows the spectrum of a 2FSK and a 2GFSK modulated signal with 40 kbps data rate and 20 kHz deviation.
3. 2(G)FSK vs. 4(G)FSK
A typical application of 4(G)FSK modulation is transmitting with the same data rate but occupying only half the bandwidth of the 2(G)FSK signal. For example, 2GFSK, 40kbps data rate, 20kHz deviation (OBW = 80 kHz) and 4GFSK, 20ksps symbol rate (= 40 kbps data rate), 10 kHz outer deviation (= 10/3 kHz inner deviation) (OBW = 40 kHz). The smaller bandwidth introduces lower noise level in the receiver, this results sensitivity improvement. On the other hand, the deviation on inner symbols will be lower, which causes sensitivity loss. The two effects compensate each other, but the latter one has a stronger effect on sensitivity, so eventually in this case 4(G)FSK will have slightly (~2 dB) worse sensitivity than 2(G)FSK.
To have the same occupied bandwidth for 4(G)FSK as 2(G)FSK, we need to use higher data rate. For example, 2GFSK, 100kbps data rate, 50kHz deviation (OBW = 200 kHz) and 4GFSK, 100ksps symbol rate (= 200 kbps data rate), 50 kHz outer deviation (= 50/3 kHz inner deviation) (OBW = 200 kHz). This way, the occupied bandwidth will be the same, but the (inner) deviation will be less for 4(G)FSK, eventually this results sensitivity degradation (~5 dB).
For more details on sensitivity difference between 2(G)FSK and 4(G)FSK modulation, refer to http://community.silabs.com/t5/Wireless-Knowledge-Base/4GFSK-vs-2GFSK-sensitivity-on-Si446x/ta-p/144255.
To sum up, 4(G)FSK is generally used to transmit with the same data rate in half the bandwidth of 2(G)FSK, or to occupy the same bandwidth using a higher data rate. In both cases 4(G)FSK will have worse sensitivity.
The above figure shows the spectrum of a 2GFSK modulated signal with 40 kbps data rate and 20 kHz deviation and a 4GFSK modulated signal with 20 ksps symbol rate and 10 kHz outer deviation.
Based on the above explained reasons, our recommendation is 2GFSK modulation, since it is more spectrally efficient than OOK or 2FSK, the Gaussian filter suppresses the spurious contents and side lobes on the transmitter side significantly while the receiver sensitivity is just slightly worse than 2FSK and a few dB better than 4GFSK.