The signal constellation for a communication system with 16 equiprobable symbols is shown as below. The channel is AWGN with noise power spectral density of $N_ 0/2.$Using the union bound, find a bound in terms of $A$ and $N_0$ on the error probability for this channel.

enter image description here

Does the union bound means error probability?the solution said $$P_e \le 15Q\left(\sqrt{\frac{d^2_{min}}{2N_0}}\right)=15Q\left(\sqrt{\frac{2A^2}{N_0}}\right) $$

but why is

\begin{align} &2\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+3\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+3\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+2\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)\\ &+3\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+4\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+4\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+3\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)\\ &+3\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+4\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+4\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+3\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)\\ &+2\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+3\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+3\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)+2\cdot\frac{1}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)\\ &=\frac{48}{16}Q\left(\frac{A}{\sqrt{\frac{N_0}{2}}}\right)\quad{?} \end{align}

enter image description here


1 Answer 1


In general, the symbol error probability is given by


where $P[s_i]$ is the probability that symbol $s_i$ was sent, $P[E|s_i]$ is the error probability given that $s_i$ was sent, and $M$ is the number of symbols. If all symbols are equally likely, $(1)$ simplifies to


For $16$-QAM it is possible to derive an exact expression for the error probabilities $P[E|s_i]$ (and hence for the symbol error probability $P_E$). You can find the derivation in most digital communication text books. A good approximation (better than the union bound) is given in this answer.

The union bound approximation gives an upper bound on $P[E|s_i]$:

$$P[E|s_i]\le\sum_{j\neq i}P[s_j\text{ chosen}|s_i\text{ sent}]\tag{3}$$

Under the given assumptions we have

$$P[s_j\text{ chosen}|s_i\text{ sent}]=Q\left(\sqrt{\frac{d^2_{ij}}{2N_0}}\right)\tag{4}$$

where $d_{ij}$ is the distance between symbols $s_i$ and $s_j$. With $(3)$ and $(4)$ the union bound approximation of $(2)$ becomes

$$P_E\le\frac{1}{M}\sum_{i=1}^M\sum_{j\neq i}Q\left(\sqrt{\frac{d^2_{ij}}{2N_0}}\right)\tag{5}$$

Note that generally you will get a sum of $Q$ functions with different arguments because the distances $d_{ij}$ are usually not all the same. This is also the case for $16$-QAM.

The solution you provided is still more conservative (hence, worse) than the union bound. That solution uses the fact that the $Q$ function is monotonically decreasing, which implies

$$Q\left(\sqrt{\frac{d^2_{ij}}{2N_0}}\right)\le Q\left(\sqrt{\frac{d^2_{min}}{2N_0}}\right)\tag{6}$$

with $d_{min}=\min_{i\neq j}\{d_{ij}\}$ being the smallest occurring distance between symbols. Plugging $(6)$ into $(5)$ we finally get the bound

$$P_E\le (M-1)Q\left(\sqrt{\frac{d^2_{min}}{2N_0}}\right)\tag{7}$$

which, as mentioned above, is even less tight than the union bound.

  • $\begingroup$ so union bound is one of method of calculating the error probability,and better than mine,but do you have the proof about $P_e \le (M-1)Q(\sqrt{\frac{d^2_{min}}{2N_0}})$ ? $\endgroup$
    – Shine Sun
    Dec 12, 2017 at 10:11
  • $\begingroup$ Is Eq. (7) a definition?if not,it must have a proof,because i don't understand why the enominator in Q function is $2N_0$,not $\frac{N_0}{2}$ $\endgroup$
    – Shine Sun
    Dec 12, 2017 at 11:45
  • $\begingroup$ @ShineSun: The noise variance is indeed $N_0/2$, and the argument of the $Q$ function must be half the distance between the symbols divided by the square root of the noise variance: $\frac{d/2}{\sqrt{N_0/2}}=\sqrt{\frac{d^2}{2N_0}}$ $\endgroup$
    – Matt L.
    Dec 12, 2017 at 12:45
  • $\begingroup$ ok! now i think my problem is (M-1),because the difference of union bound and mine is the value before Q function.How does (M-1) come from? $\endgroup$
    – Shine Sun
    Dec 12, 2017 at 13:15
  • $\begingroup$ @ShineSun: Just plug (6) into (5) and note that the sum $\sum_{j\neq i}$ is over $M-1$ equal numbers. $\endgroup$
    – Matt L.
    Dec 12, 2017 at 13:49

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.