bladeRF with Linrad on 432 MHz.
(Jan 2 2014)

The bladeRF hardware.

The bladeRF hardware is available from NUAND The unit uses a chip from Lime Microsystems, the LMS6002D. The nominal frequency coverage is 300 MHz to 3.8 GHz. The unit can also be used from DC to perhaps 18 MHz as a two channel HF receiver.

The hardware is a transceiver. This page is only about how the receiver is implemented in Linrad and what performance data is observed when the tuner frequency is set to 432 MHz.

The bladeRF can supply 12 bit I/Q data at a sampling speed of 40 MHz. Linrad can not process more than about 10 MHz of bandwidth today (year 2013) on a Xeon E5410 CPU at 2.33 GHz. (Performance is limited by the first FFT which can only use a single CPU core.) It is possible to set the bladeRF to send data at 40 MHz and to set Linrad to average samples four by four. That procedure reduces the A/D noise floor by 6 dB but since the noise from the tuner is dominating the improvement by running the A/D four times faster is very small. The LMS6002 chip is not designed to be used for recieving a weak signal within a passband containing very strong signals. It is primarily intended to receive wideband transmissions where the fairly sharp filters in the LMS6002D are matched to the bandwidth of the desired signal. The LMS6002D seems to have a little too much gain. Even at minimum gain there is a significan amount of noise in the ADC even when the LNA is disabled. For optimum dynamic range in an amateur radio system one would like the noise with LNA off to be below a single bit and then with LNA enabled one would want a noise floor with an RMS value of something like 3 bits.

For the use in amateur radio, the excessive gain of the LMS6002 is totally irrelevant because at spacings below about 500 kHz the noise floor is determined by reciprocal mixing and there is no advantage in making the noise floor lower. There could be other applications such as FM DX-ing where it could make a difference.

In Linrad the user can choose between sensitivity mode and linearity mode. They differ in that gain reduction in sensitivity mode is done late in the signal path while in linearity mode it is done in the input stage and in the mixer. If the bladeRF is used with a mast mounted pre-amplifier and a selective filter the sensitivity mode will give a good dynamic range.

When the bladeRF is run directly from an antenna it may be better to turn down the front end gain to avoid saturating the LNA with signals far from the frequency of interest.

Linearity problems.

Figure 1 shows the spectrum with bladeRF sampling at 40 MHz and with Linrad operating on averages of four samples. The signal level is -30 dBm. About 2 dB from saturation.

Fig. 1. A signal on 432.010 MHz is injeted into the bladeRF. The generator indicates -30 dBm but the real level is a little lower. The signal is about 2 dB from saturation. The center frequency is 431.75 so the fundamental in the baseband is 260 kHz. High order harmonics of 260 kHz have an appearent level of about -70 dBm to -80 dBm.

It is obvious that there is a linearity problem. Plenty of high order harmonics occur only 40 to 50 dB below the main signal. The input level has to be drastically reduced for the high order harmonics to disappear. First at -80 dBm they are no longer visible. For details look here. The high order harmonics are probably caused by the A/D converters of by feedback from the digital side to the input of the baseband amplifiers.

When using the bladeRF as the receiver in an amateur station on 432 MHz one could sample at 40 MHz and process 10 MHz in Linrad while having a much narrower system bandwidth defined by a filter on 432 MHz. By placing the signal well separated from the center one would eliminate interference due to overtones from signals nearer the center. They would be rejected by the filter.

If there are no strong signals that could generate overtone interferences one could allow Linrad to use much lower bandwidth. Note that one has to select a filter to avoid aliases. The narrowest filter is 1.5 MHz in the current bladeRF library (Dec. 2013) so the lowest useful fft1 bandwidth is 2 MHz.

Reciprocal mixing.

Figure 2 shows the screen with bladeRF sampling at 20 MHz with Linrad operating on averages of four samples giving a bandwidth of 5 MHz. The filter in the LMS6002D is set to 3 MHz and the gain mode is set to linearity.

Fig. 2. A signal on 432.010 MHz is injeted into the bladeRF. The S-meter is calibrated to show 0 dBm for the signal. When the frequency is stepped by 20 kHz the level changes by 68.8 dB in a bandwidth of 1 kHz. This means that reciprocal mixing is at 98.8 dBc/Hz at a frequency separation of 20 kHz. That means that the dynamic range in 500 Hz bandwidth is a mediocre 71.8 dB at 20 kHz frequency separation.

Reciprocal mixing at other frequency separations than 20 kHz can be estimated from figure 2 and at large frequency separations from figure 1.


The frequency stability of the bladeRF is good. Figure 3 shows the reception of two signals. The baseband waterfall spans about 6 minutes. The signal on 432100.079 kHz is from a HP8644A with the low sideband noise option. The signal on 432100.098 is from a HP8657A which was locked to a good external 10 MHz standard.

Fig. 3. This high resolution waterfall shows two independent signal generators. It is obvious that the stability of the bladerf is comparable to the stability of the generators. see text.

It is obvious from figure 3 that the random frequency fluctuations of the two signals are different to a significant degree which means that the frequency stability of the bladeRF is comparable to the frequency stability of the generators. The baseband spectrum shows a significant broadening of the spectral lines. The reason could be mainly in the LMS6002D tuner, but it could also be due to the signal generators. I see no reason to investigate this further because the stability is adequate for all usages I am aware of.

Noise figure and dynamic range.

A HP8657A was used to send a weak signal into the bladeRF and the level was adjusted for the noise floor in a bandwidth of 10 kHz to rise by 3.0 dB. The noise floor in 1 Hz is then 40 dB below the value one can read on the signal generator. On an ideal receiver with NF=0dB one would then read -134 dBm.

To calibrate the signal generator two different wideband amplifiers were used. PSA4-4053 which has NF=0.62 dB on 144 MHz and AD6IW which has NF=0.47 dB on 144 MHz. (Details about how these NF values were established can be found here. Precision Measurements of noise figures

To get the NF of the two amplifiers on 1296, they were both compared to a L LNA for which the NF is known to be 0.27 dB on 1296 MHz. comparing three different noise heads

The S/N for a fairly strong signal for the three amplifiers on 1296 MHz was measured and these values were obtained:

LLNA 21.43 dB
AD6IW 21.05 dB
PSA4-4053 20.62 dB

From this one can conclude the NF of AD6IW is 0.65 dB on 1296 MHz and that the NF of the PSA4-4053 is 1.08 dB on 1296 MHz. For these wideband amplifiers it is reasonable to assume that the NF on 432 is the average of the NF at 144 and 1296 which means one can assume this:

AD6IW NF = 0.56 dB @ 432 MHz.
PSA4-4053 NF = 0.85 dB @ 432 MHz.

The signal level as indicated on the HP8657A required to produce S/N=3 dB in a bandwidth of 10 kHz is -133.0 dBm for the AD6IW and -132.7 dBm for the PSA4-4053. From those values one would compute the NF to 1 dB and 1.3 dB respectively. The calibration error is thus 0.45 dB on the power level of the HP8657A. (The true power is 0.45 dB lower than indicated on the generator.)

Table 1 shows the noise floor in dBm/Hz at 432.100 MHz as well as the signal level on 432.120 and 900.0 MHz required to degrade S/N by 3 dB for a weak signal on 432.1 MHz.

           Gain mode sensitivity                     Gain mode Linearity
Gain    Noise  NF     432.12          900         Noise   NF     432.12        900       
(dB)  (dBm/Hz)(dB)  (dBm)  (dB)  (dBm)   (dB)    (dBm/Hz)(dB)  (dBm) (dB)  (dBm)   (dB) 
 21    -162.1 11.9  -62.1  73.0  -24.6rc 114.5    -161.1 12.9  -62.9 71.1  -23.6c  110.5
 18    -162.2 11.8  -62.7  72.5  -24.6rc 114.6    -159.1 14.9  -60.5 71.6  -21.6r  110.5
 15    -162.3 11.7  -62.7  72.6  -24.0rc 115.3    -156.9 17.1  -58.5 71.4  -21.6rc 108.3
 12    -162.0 12.0  -62.3  72.7  -24.0rc 115.0    -154.5 19.5  -56.1 71.4  -19.0c  108.5
  9    -161.5 12.5  -63.3  71.2  -22.6c  115.9    -149.2 24.8  -50.1 72.1   -9.2r  113.0
  6    -160.4 13.6  -60.9  72.5  -22.6c  114.8    -146.7 27.3  -48.1 71.6   -7.2r  112.5
  3    -158.6 15.4  -60.9  70.7  -22.4r  113.2    -146.7 27.3  -47.7 72.0   -7.2r  112.5
  0    -155.6 18.4  -58.9  69.7  -22.4r  110.2    -146.7 27.3  -47.7 72.0   -7.2r  112.5
 -3    -153.2 20.8  -56.7  69.5  -22.0c  108.2    -145.1 28.9  -45.2 72.9   -5.6r  112.5
 -6    -150.7 23.3  -51.5  72.2  -17.4r  110.3    -141.8 32.2  -43.5 71.3   -3.6r  111.2
 -9    -147.7 26.3  -50.7  70.0  -17.4r  107.3    -139.1 34.9  -40.3 71.8   -1.8r  110.3
-12    -144.3 29.7  -47.7  69.6  -17.0c  104.3    -136.2 37.8  -37.7 71.5    0.0c  109.2
-15    -141.8 32.2  -44.7  70.1   -4.6c  114.2    -133.3 40.7  -34.7 71.6    1.0c  107.3
-18    -139.0 35.0  -41.3  70.7   -2.0rc 114.0    -132.5 41.5  -33.7 71.8    1.0c  107.3
-21    -136.2 37.8  -35.1  74.1   -1.0c  112.2    -132.5 41.5  -33.7 71.8    1.0c  107.3

Table 1. The noise floor and the dynamic range at close and wide spacing for the bladeRF. The dB column for dynamic range is normalized to a bandwidth of 500 Hz. The letters r and c in the dynamic range dBm columns indicate whether the limitation looks mostly like compresson (c) or like reciprocal mixing(r).

Table 1 is not normal. The dynamic range is limited by an increase of the noise floor at wide frequency separations. Usually that means reciprocal mixing, but the effect changes depending on the gain settings in an irregular way and that indicates that the variable gain amplifiers add noise modulation. (most probably on the amplitude.)

There is not much difference between the gain modes. (As there should be.)

Use this radio with a proper LNA and a selective filter. It should be a good tool to use on the higher amateur bands providing a wide frequency coverage, good frequency stability and a close range dynamic range of about 70 dB in 500 Hz. The wide range dynamic range (300 kHz) is about 90 dB as can be seen in figure 2. For signals outside the filter of the LMS6002D, the dynamic range is in the order of 110 dB in a bandwidth of 500 Hz regardless of the frequency separation.

Two tone test.

Figures 4 to 7 show the screen with two equally strong signals. The signal frequencies are 432.0 and 432.1. In figure 4 the signal level for each tone is -55 dBm and the two tones together are less then 1 dB from saturation. A large number of false signals are present at levels about 50 dB below the signals. They are separated by 100 kHz. Note the strong signal at 100 kHz. It is the difference frequency, probably IM2, from the tuner and as expected the sum frequency at 433.1 MHz has a similar level.

The false frequencies depend in a regular way on the signal level so they are not caused by digital feedback, which is a serious problem with the bladeRF when used for HF on the J61 connector.

Fig. 4. A two-tone test near the point of saturation. Two tones at -55 dBm.

Fig. 5. A two-tone test 10 dB below the point of saturation. Two tones at -65 dBm.

Fig. 6. A two-tone test 20 dB below the point of saturation. Two tones at -75 dBm.

Fig. 7. A two-tone test 30 dB below the point of saturation. Two tones at -85 dBm.