I read the article Google wants the US' wireless spectrum for balloon-based Internet. It says to use over 24 GHz frequency spectrum for communication.

Is it ever possible to generate that high frequency by using piezoelectric crystals? Or are they using a PLL frequency multiplier?

Even if it is possible to generate that high-frequency signal, and if you want to send 1 bit on every period of signal, there must be a processor that is working much more faster than 24 GHz. How is that possible on a balloon?

  • 12
    \$\begingroup\$ 24GHz is the proposed RF carrier frequency, not the signal bandwidth, nor the bit rate. (News media seldom understand technical details.) The article is about Google requesting regulatory approval, which is just the first step to legal operation. Article doesn't seem to detail what kind of modulation they intend to use. \$\endgroup\$
    – MarkU
    Commented Feb 6, 2015 at 8:32
  • 1
    \$\begingroup\$ @Arsenal Usually it's germanium or silicon/germanium that is used in high frequency applications like that - not difficult to make small chips that work well into the 10s of GHz. \$\endgroup\$
    – J...
    Commented Feb 6, 2015 at 12:56
  • 7
    \$\begingroup\$ It may be worth mentioning that while we don't think of it in these terms, visible light is e.g. 590 THz for green. \$\endgroup\$
    – Random832
    Commented Feb 6, 2015 at 13:26
  • 6
    \$\begingroup\$ Um, you do realize that you can do amplitude modulation of most signals between Mhz and THz (Tera Hertz) signal with nothing more but your hand, right? As in: wave hand in front of the antenna/waveguide/light source. So, if your naked body can pull that off, it's not surprising that you can do it with a bit of electronics, too :) This also brings into focus the fact that you don't need mechanical oscillation to produce a frequency reference. You can have the bound electrons, or individual atoms or molecules oscillate, too! \$\endgroup\$ Commented Feb 6, 2015 at 14:53
  • 1
    \$\begingroup\$ You don't have to generate the carrier frequency directly from a crystal. You can use a PLL to get your 24 GHz from a much lower reference frequency. \$\endgroup\$
    – Austin
    Commented Feb 6, 2015 at 17:18

3 Answers 3


RF comms do not transmit one bit of information per cycle of the carrier wave - that would be digital baseband communications and it requires incredible amounts of bandwidth. Incidentally, you can buy FPGAs with built-in 28 Gbps serdes hard blocks. These can serialize and deserialize data for 100G ethernet (4x25G + coding overhead). I suppose the 'fundamental' frequency in this case would actually be 14 GHz (data rate/2 - think about why this is!) and they require around 200 MHz to 14 GHz of bandwidth. They don't go all the way down to DC due to using the 64b66b line code. The frequency used to drive the serdes modules would be generated by some sort of a VCO that is phase locked to a crystal reference oscillator.

In the RF world, the message signal is modulated onto a carrier which is then upconverted to the required frequency for transmission with mixers. These balloons probably have a baseband of less than 100 MHz, meaning that initially the digital data is modulated onto a relatively low frequency carrier (intermediate frequency) of around 100 MHz. This modulation can be done digitally and the modulated IF generated by a high speed DAC. Then this frequency is translated up to 24 GHz with a 23.9 GHz oscillator and a mixer. The resulting signal will extend from 23.95 to 24.05 GHz, 100 MHz of bandwidth.

There are many ways to build high frequency oscillators in that band. One method is to build a DRO, which is a dielectric resonance oscillator. Think of this as an LC tank circuit - there will be some frequency where it will 'resonate' and either generate a very high or very low impedance. You could also think of this as a narrow bandpass filter. In a DRO, a piece of dielectric is used - usually some sort of ceramic, I believe - that resonates at the frequency of interest. The physical size and shape determine the frequency. All you need to do to turn it into a frequency source is add some gain. There are also ways of using special diodes that exhibit negative resistance. A Gunn diode is one example. Biasing a Gunn diode correctly will cause it to oscillate at several GHz. Another possibility is something called a YIG oscillator. YIG stands for Yttrium Iron Garnet. It is common to build bandpass filters by taking a small YIG sphere and coupling it to a pair of transmission lines. YIG happens to be sensitive to magnetic fields, so you can tune or sweep the center frequency of the filter by varying the ambient magnetic field. Add an amplifier, and you have a tunable oscillator. It's relatively easy to put a YIG in a PLL. The power of a YIG is that it is possible to use it to produce a very wide band smooth sweep, and hence they are often used in RF test equipment such as spectrum and network analyzers and sweeping and CW RF sources. Another method is to simply use a bunch of frequency multipliers. Any nonlinear element (such as a diode) will produce frequency components at multiples of the input frequency (2x, 3x, 4x, 5x, etc.). Stringing together a chain of multipliers, bandpass filters, and amplifiers can be used to produce very high frequencies.

  • 5
    \$\begingroup\$ Can you provide a layman's summary? This answer is 100% technobabble! \$\endgroup\$ Commented Feb 6, 2015 at 14:04
  • 6
    \$\begingroup\$ @LightnessRacesinOrbit TL; DR: 1) 24GHz signaling frequency does not spell 24Gbaud; 2) The 24GHz RF can be generated using a much lower frequency signal that a processor can handle (e.g 100MHz directly off a fast DAC), a constant high frequency feed and a mixer (like those 6-transistor superheterodyne radio); 3) a multi-gigahertz oscillator is very easy to build now, with multiple possible ways. \$\endgroup\$ Commented Feb 6, 2015 at 16:38
  • \$\begingroup\$ @MaxthonChan: I meant in the answer :) \$\endgroup\$ Commented Feb 6, 2015 at 16:39
  • \$\begingroup\$ @LightnessRacesinOrbit This is my attempt at writing a layman's summary, hence I prefixed it with a "TL; DR" in bold fonts. \$\endgroup\$ Commented Feb 6, 2015 at 17:09
  • \$\begingroup\$ @Max Yes I get that and I appreciate it. I'm suggesting that it be inserted into the answer as comments are transient. Cheers \$\endgroup\$ Commented Feb 6, 2015 at 20:40

Here's my attempt at a layman summary, adapted from this answer.

When we talk about communication happening "at 24 GHz", we're referring to a small range of frequencies. In order for the signal "at 24 GHz" not to trample all over the signals at all the other frequencies, there's a hard limit on how much the signal is allowed to differ from a 24 GHz sinewave.

The whole point of having a radio "band" is that by placing a limit on how much the signal can differ to a sinewave, it becomes possible to create filters which remove signals that differ too much from your sinewave, thus suppressing them and keeping only the signal you're interested in.

For example, here is random noise filtered to contain only frequencies between 190 Hz and 210 Hz:

enter image description here

Observe that it is not that far off from a (200 Hz) sinewave. For comparison, here's noise filtered to contain 150 Hz to 250 Hz:

enter image description here

Note how it differs much more from a perfect sinewave. Now, if you take a 24 GHz sinewave and start arbitrarily turning bits of it on and off, the receiver will not see it the way you send it, because turning bits on/off arbitrarily will make the signal fall outside of the 24 GHz range. The receiver will filter out frequencies outside the 24 GHz range, thus distorting the signal. Bottom line is: if you modulate the signal naively by turning bits on and off, it won't work with the idea of filtering out unwanted frequencies.

Before filtering, the above signal looked like this:

enter image description here

Think of it as what a radio receiver sees before it filters out unwanted frequencies. I think that's a reasonable layman approximation. Note that the horizontal scale here is exactly the same as in the images above – what you're seeing are all the frequencies higher than 200-odd Hz. Frequencies below 200 Hz are also there, but they are not obvious to the naked eye.

(the math works the same at Hz or GHz scales, so don't let this put you off)

  • 1
    \$\begingroup\$ To an RF layman like me, this is an EXCELLENT answer. What equation(s) describe the hard limit? \$\endgroup\$ Commented Feb 8, 2015 at 2:26
  • 2
    \$\begingroup\$ @BenSimmons the hard limit is actually up to the RF designer to choose, and the trade-off is how much of the frequency spectrum your signal "eats up" and takes away from other uses, vs. how much information one can carry for a given signal-to-noise ratio. See Shannon-Hartley theorem. So a high bandwidth means you allow the signal to differ from your 24 GHz sinewave a lot, and a low bandwidth = smaller differences are allowed. \$\endgroup\$ Commented Feb 8, 2015 at 13:47
  • \$\begingroup\$ Interesting. Is the noise power fairly constant everywhere? I'm just wondering how the signal power is decided upon. Is it ever "adaptive" to the environment, for instance is noise level changes? \$\endgroup\$ Commented Feb 9, 2015 at 8:20
  • 2
    \$\begingroup\$ @BenSimmons the RF noise is definitely not constant; much noise is produced by human-made transmitters because perfect transmission is impossible, but solar activity etc makes RF noise too. Some noise is not received but rather added by the receiver amplifiers etc. I believe Wi-Fi a/b/g usually transmits at max power possible, to achieve best signal-to-noise ratio, while cell phones vary transmit power to save battery (don't quote me on this!...). Cell towers, TV towers etc broadcast to many receivers and so cannot really adjust power based on any sort of feedback. \$\endgroup\$ Commented Feb 9, 2015 at 15:14
  • \$\begingroup\$ The cell phone tower commands the phone transmit power level, and this is updated continuously to maintain a constant SNR. This is called 'closed loop power control'. This is required not only to minimize power consumption but also as a result of CDMA coding. Since the base station is a single antenna, it can use orthogonal codes that do not interfere with each other. However, it is not possible to achieve the required synchronization to use orthogonal codes the other way, so the cell phone signals interfere with each other and the transmit power must be controlled to minimize this. \$\endgroup\$ Commented Feb 10, 2015 at 8:12

FM radio transmits on a 98MHz +-10MHz carrier frequency, but each station only has about 200khz worth of information (occupied bandwidth). Similarly, DirecTV transmits on a 14GHz carrier frequency, but the signal is probably only 10 or 100's of MHz of occupied bandwidth.

Presumably, Google wants to use the 24GHz band to carry signals with much lower occupied bandwidth. But if someone wanted to actually transmit such a large amount of bandwidth, it can be done, by various modulation techniques using multiple carriers.

As far as the actual electronics, I have seen 24GHz MMICs before. Also, you are presuming that a single "processor" is necessary. You could have 24 1Gbit/second modems stacked doing FDMA. The 100Gb/sec ethernet that Xilinx is capable of, as discussed above, I think uses parallel Quad GMII interfaces.

The EM spectra is a continuum, and as you increase frequency, eventually you go from RF to optical. There are line-of-sight Laser Comm systems in existence.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.